diff --git a/.gitignore b/.gitignore
index 0ef476e5..803c29b2 100644
--- a/.gitignore
+++ b/.gitignore
@@ -47,4 +47,4 @@ docs/_build/
 
 # Keras model files
 history.p
-keras_model.*
\ No newline at end of file
+keras_model*
\ No newline at end of file
diff --git a/README.md b/README.md
index f8eb5090..dfdd432c 100644
--- a/README.md
+++ b/README.md
@@ -61,19 +61,6 @@ knowledgehub -c /etc/ckan/default/production.ini db init
 sudo service apache2 reload
 ```
 
-6. Run the command for predictive search periodically( daily or weekly is recommended ).
-This command will start training the model:
-
-```
-knowledgehub  -c /etc/ckan/default/production.ini predictive_search train
-```
-
-There is a action that can run CLI commands for Knowledge Hub.
-This example shows how to run the above command through the API action:
-```
-curl -v 'http://hostname/api/3/action/run_command' -H'Authorization: API-KEY' -d '{"command": "predictive_search train"}'
-```
-
 ### Config Settings
 
 These are the required configuration options used by the extension:
@@ -94,24 +81,24 @@ ckanext.knowledgehub.sub_themes_per_page = 20
 ckanext.knowledgehub.dashboards_per_page = 20
 ```
 4. Predictive Search
-     - Length of the seuqunce after which the model can start predict, recommended at least 15 chars long
+     - Length of the sequence after which the model can start predict, recommended at least 10 chars long
      ```
      # (optional, default: 10)
      ckanext.knowledgehub.rnn.sequence_length = 12
      ```
      - Number of chars to be skipped in generation of next sentence
      ```
-     # (optional, default: 3)
+     # (optional, default: 1)
      ckanext.knowledgehub.rnn.sentence_step = 2
      ```
      - Number of predictions to return
      ```
      # (optional, default: 3)
-     ckanext.knowledgehub.rnn.number_prediction = 2
+     ckanext.knowledgehub.rnn.number_predictions = 2
      ```
      - Minimum length of the corpus after it should start to predict
      ```
-     # (optional, default: 3)
+     # (optional, default: 10000)
      ckanext.knowledgehub.rnn.min_length_corpus = 300
      ```
      - Maximum epochs to learn
@@ -119,10 +106,15 @@ ckanext.knowledgehub.dashboards_per_page = 20
      # (optional, default: 50)
      ckanext.knowledgehub.rnn.max_epochs = 30
      ```
-     - Full path to the RNN model
+     - Full path to the RNN weights model
+     ```
+     # (optional, default: ./keras_model_weights.h5)
+     ckanext.knowledgehub.rnn.model_weights = /home/user/model_weights.h5
      ```
-     # (optional, default: ./keras_model.h5)
-     ckanext.knowledgehub.rnn.model = /home/user/model.h5
+     - Full path to the RNN network model
+     ```
+     # (optional, default: ./keras_model_network.h5)
+     ckanext.knowledgehub.rnn.model_network = /home/user/model_network.h5
      ```
      - Full path to the model history
      ```
@@ -226,7 +218,7 @@ knowledgehub -c /etc/ckan/default/production.ini search-index rebuild --model da
 
 This would rebuild the index for dashboards.
 
-Avalilable model types are: 
+Avalilable model types are:
 * `ckan` - rebuilds the CKAN core (package) index,
 * `dashboard` - rebuilds the dasboards index,
 * `research-question` - rebuilds the research questions index and
@@ -264,4 +256,24 @@ The crontab should look something like this:
 
 Data Quality is measured across the six primary dimensions for data quality assessment.
 
-A lot more details are available in the dedicated [documentation section](docs/data-qualtiy-metrics.md).
\ No newline at end of file
+A lot more details are available in the dedicated [documentation section](docs/data-qualtiy-metrics.md).
+
+# Predictive search
+
+The preditive search functinality predict the next n characters in the word or the next most possible word.
+The training data is consist of title and description of all entities on Knowledge hub including themes, sub-themes,
+research questions, datasets, visualizations and dashboards. Before it starts predict the machine learning model has to be trained.
+By default user should write 10 characters in the search box on home page before it starts to predict.
+
+Run the command for predictive search periodically( daily or weekly is recommended ).
+This command will start training the model:
+
+```
+knowledgehub  -c /etc/ckan/default/production.ini predictive_search train
+```
+
+There is a action that can run CLI commands for Knowledge Hub.
+This example shows how to run the above command through the API action:
+```
+curl -v 'http://hostname/api/3/action/run_command' -H'Authorization: API-KEY' -d '{"command": "predictive_search train"}'
+```
\ No newline at end of file
diff --git a/ckanext/knowledgehub/cli/predictive_search.py b/ckanext/knowledgehub/cli/predictive_search.py
index abe7d976..1c66f965 100644
--- a/ckanext/knowledgehub/cli/predictive_search.py
+++ b/ckanext/knowledgehub/cli/predictive_search.py
@@ -4,7 +4,7 @@
 import logging
 
 from ckanext.knowledgehub.cli import error_shout
-from ckanext.knowledgehub.rnn.worker import learn
+from ckanext.knowledgehub.lib.rnn import PredictiveSearchWorker
 
 log = logging.getLogger(__name__)
 
@@ -19,7 +19,8 @@ def train():
     u'''Initialising the Knowledgehub tables'''
     log.info(u"Initialize Knowledgehub tables")
     try:
-        learn()
+        worker = PredictiveSearchWorker()
+        worker.run()
     except Exception as e:
         error_shout(e)
     else:
diff --git a/ckanext/knowledgehub/fanstatic/javascript/search_prediction.js b/ckanext/knowledgehub/fanstatic/javascript/search_prediction.js
index 5ac7033c..40fea859 100644
--- a/ckanext/knowledgehub/fanstatic/javascript/search_prediction.js
+++ b/ckanext/knowledgehub/fanstatic/javascript/search_prediction.js
@@ -50,6 +50,7 @@
                 x[i].parentNode.removeChild(x[i]);
             }
         }
+        currentFocus = -1;
     }
 
     $(document).ready(function () {
@@ -58,21 +59,21 @@
         searchInput
             .bind("change keyup", function (event) {
                 clearTimeout(timer)
-                if (!(event.keyCode >= 13 && event.keyCode <= 20) && !(event.keyCode >= 37 && event.keyCode <= 40)) {
+                if (!(event.keyCode >= 13 && event.keyCode <= 20) && !(event.keyCode >= 37 && event.keyCode <= 40) && event.keyCode != 27) {
                     // detect that user has stopped typing for a while
                     timer = setTimeout(function() {
                         var text = searchInput.val();
 
                         if (text !== '') {
                             api.get('get_predictions', {
-                                text: text
+                                query: text
                             }, true)
                             .done(function (data) {
                                 if (data.success) {
                                     var a, b;
                                     var results = data.result;
 
-                                    closeAllLists()
+                                    closeAllLists();
 
                                     a = document.createElement("DIV");
                                     a.setAttribute("id", "autocomplete-list");
@@ -84,8 +85,9 @@
                                         b.innerHTML = text;
                                         b.innerHTML += "<strong>" + r + "</strong>";
                                         b.addEventListener("click", function (e) {
-                                            searchInput.val(text + r);
                                             closeAllLists();
+                                            searchInput.val(text + r);
+                                            searchInput.trigger("change");
                                         });
                                         a.append(b)
                                     });
@@ -95,25 +97,24 @@
                                 console.log("Get predictions: " + error.statusText);
                             });
                         }
-                    }, 500);
+                    }, 300);
                 }
             })
     });
 
     $('.search-input-group').on("mouseover", autocompleteItems, function(e){
-
         var activeItem = document.getElementsByClassName('autocomplete-active')[0];
         activeItem ? activeItem.classList.remove('autocomplete-active') : null;
         event.target !== input ? event.target.classList.add('autocomplete-active') : null;
         var p = e.target.parentElement;
         var index = Array.prototype.indexOf.call(p.children, e.target);
         activeItem ? currentFocus = index : currentFocus = -1
-
     });
 
     searchInput.on('keydown', function (e) {
         var x = document.getElementById("autocomplete-list");
         if (x) x = x.getElementsByTagName("div");
+
         if (e.keyCode == 40) {
             // The arrow DOWN key is pressed
             currentFocus++;
@@ -125,9 +126,12 @@
         } else if (e.keyCode == 13) {
             // ENTER key is pressed
             if (currentFocus > -1) {
+                e.preventDefault();
                 // simulate a click on the "active" item*
                 if (x) x[currentFocus].click();
             }
+        } else if (e.keyCode == 27) {
+            closeAllLists();
         }
     });
 })(ckan.i18n.ngettext, $);
\ No newline at end of file
diff --git a/ckanext/knowledgehub/fanstatic/javascript/user_query_result.js b/ckanext/knowledgehub/fanstatic/javascript/user_query_result.js
index 68693a17..9c76a4aa 100644
--- a/ckanext/knowledgehub/fanstatic/javascript/user_query_result.js
+++ b/ckanext/knowledgehub/fanstatic/javascript/user_query_result.js
@@ -37,10 +37,10 @@
                     result_id: result_id
                 })
                 .done(function (data) {
-                    console.log("User Quere Result: SAVED!");
+                    console.log("User Query Result: SAVED!");
                 })
                 .fail(function (error) {
-                    console.log("User Quere Result failed: " + error.statusText);
+                    console.log("User Query Result failed: " + error.statusText);
                 });
             }
         })
@@ -49,6 +49,19 @@
         });
     }
 
+    function saveKnowledgeHubData(query_text) {
+        api.post('kwh_data_create', {
+            type: 'search_query',
+            title: query_text
+        })
+        .done(function (data) {
+            console.log("User query added to kwh data");
+        })
+        .fail(function (error) {
+            console.log("Failed to add user query to kwh data: " + error.statusText);
+        });
+    }
+
     $(document).ready(function () {
         var save_user_query = function(callback) {
             var tab_content = $('.tab_content');
@@ -81,6 +94,7 @@
                 if (query_text) {
                     var user_id = $('#user-id').val();
                     saveUserQueryResult(query_text, result_type, result_id, user_id)
+                    saveKnowledgeHubData(query_text, user_id)
                 }
             }
         });
diff --git a/ckanext/knowledgehub/helpers.py b/ckanext/knowledgehub/helpers.py
index c38b596e..2d164c64 100644
--- a/ckanext/knowledgehub/helpers.py
+++ b/ckanext/knowledgehub/helpers.py
@@ -31,7 +31,6 @@
 
 from ckanext.knowledgehub.model import Dashboard
 from ckanext.knowledgehub.model import ResourceValidation
-from ckanext.knowledgehub.rnn import helpers as rnn_helpers
 
 
 log = logging.getLogger(__name__)
diff --git a/ckanext/knowledgehub/lib/rnn/__init__.py b/ckanext/knowledgehub/lib/rnn/__init__.py
new file mode 100644
index 00000000..f6cc90e2
--- /dev/null
+++ b/ckanext/knowledgehub/lib/rnn/__init__.py
@@ -0,0 +1,8 @@
+from ckanext.knowledgehub.lib.rnn.worker import PredictiveSearchWorker
+from ckanext.knowledgehub.lib.rnn.model import PredictiveSearchModel
+
+
+__all__ = [
+    'PredictiveSearchWorker',
+    'PredictiveSearchModel'
+]
diff --git a/ckanext/knowledgehub/lib/rnn/config.py b/ckanext/knowledgehub/lib/rnn/config.py
new file mode 100644
index 00000000..87b7d96d
--- /dev/null
+++ b/ckanext/knowledgehub/lib/rnn/config.py
@@ -0,0 +1,40 @@
+import os
+import time
+
+from ckan.common import config
+
+
+class PredictiveSearchConfig(object):
+    ''' Hold the configuration for the machine learning model and worker '''
+
+    def __init__(self):
+        self.corpus_length = int(config.get(
+            u'ckanext.knowledgehub.rnn.min_length_corpus', 10000))
+        self.sequence_length = config.get(
+            u'ckanext.knowledgehub.rnn.sequence_length', 10)
+        self.step = int(
+            config.get(u'ckanext.knowledgehub.rnn.sentence_step', 1))
+        self.epochs = int(
+            config.get(u'ckanext.knowledgehub.rnn.max_epochs', 50))
+        self.weights_path = config.get(
+            u'ckanext.knowledgehub.rnn.model_weights',
+            './keras_model_weights.h5'
+        )
+        self.network_path = config.get(
+            u'ckanext.knowledgehub.rnn.model_network',
+            './keras_model_network.h5'
+        )
+        self.history_path = config.get(
+            u'ckanext.knowledgehub.rnn.history',
+            './history.p'
+        )
+        self.temp_weigths_path = os.path.join(
+            os.path.dirname(self.weights_path),
+            'keras_model_%s.h5' % time.time()
+        )
+        self.number_predictions = int(
+            config.get(
+                u'ckanext.knowledgehub.rnn.number_predictions',
+                3
+            )
+        )
diff --git a/ckanext/knowledgehub/lib/rnn/data_manager.py b/ckanext/knowledgehub/lib/rnn/data_manager.py
new file mode 100644
index 00000000..726d4396
--- /dev/null
+++ b/ckanext/knowledgehub/lib/rnn/data_manager.py
@@ -0,0 +1,57 @@
+import ckan.plugins.toolkit as toolkit
+
+
+class DataManager:
+    ''' Manage the training data'''
+
+    @staticmethod
+    def create_corpus(corpus):
+        ''' Store the machine learning corpus
+
+        :param corpus: the machine learning corpus
+        :type corpus: string
+
+        :returns: the stored corpus
+        :rtype: dict
+        '''
+        return toolkit.get_action('corpus_create')(
+            {'ignore_auth': True},
+            {'corpus': corpus}
+        )
+
+    @staticmethod
+    def get_corpus():
+        ''' Get the data in knowledgehub and create corpus
+
+        :returns: the machine learning corpus
+        :rtype: string
+        '''
+        kwh_data = toolkit.get_action(
+            'kwh_data_list')({'ignore_auth': True}, {})
+
+        corpus = ''
+        if kwh_data.get('total'):
+            data = kwh_data.get('data', [])
+            for entry in data:
+                corpus += ' %s' % entry.get('title')
+                if entry.get('description'):
+                    corpus += ' %s' % entry.get('description')
+
+        return corpus
+
+    @staticmethod
+    def get_last_corpus():
+        ''' Return the corpus usd in the last training of the model '''
+
+        return toolkit.get_action('get_last_rnn_corpus')(
+            {'ignore_auth': True}, {})
+
+    @staticmethod
+    def prepare_corpus(corpus):
+        ''' Find the unique chars in the corpus and index the characters '''
+
+        unique_chars = sorted(list(set(corpus)))
+        char_indices = dict((c, i) for i, c in enumerate(unique_chars))
+        indices_char = dict((i, c) for i, c in enumerate(unique_chars))
+
+        return (unique_chars, char_indices, indices_char)
diff --git a/ckanext/knowledgehub/lib/rnn/model.py b/ckanext/knowledgehub/lib/rnn/model.py
new file mode 100644
index 00000000..20ee913a
--- /dev/null
+++ b/ckanext/knowledgehub/lib/rnn/model.py
@@ -0,0 +1,109 @@
+import logging
+import os
+
+import heapq
+import numpy as np
+import tensorflow as tf
+from keras.models import load_model, model_from_json
+from filelock import FileLock
+
+from ckanext.knowledgehub.lib.rnn.data_manager import DataManager
+from ckanext.knowledgehub.lib.rnn.config import PredictiveSearchConfig
+
+
+class PredictiveSearchModel(PredictiveSearchConfig):
+    ''' Use the machine learning model trained by the worker.
+
+    Has ability to predict the next characters or word for given text.
+    '''
+
+    def __init__(self):
+        super(PredictiveSearchModel, self).__init__()
+        self.model = None
+        self.unique_chars = None
+        self.char_indices = None
+        self.indices_char = None
+        self.logger = logging.getLogger('ckanext.PredictiveSearchModel')
+
+    def prepare_input(self, text):
+        x = np.zeros((1, self.sequence_length, len(self.unique_chars)))
+        for t, char in enumerate(text):
+            x[0, t, self.char_indices[char]] = 1.
+
+        return x
+
+    def sample(self, preds, top_n=3):
+        preds = np.asarray(preds).astype('float64')
+        preds = np.log(preds)
+        exp_preds = np.exp(preds)
+        preds = exp_preds / np.sum(exp_preds)
+
+        return heapq.nlargest(top_n, range(len(preds)), preds.take)
+
+    def predict_completion(self, text):
+        completion = ''
+        while True:
+            x = self.prepare_input(text)
+            preds = self.model.predict(x, verbose=0)[0]
+            next_index = self.sample(preds, top_n=1)[0]
+            next_char = self.indices_char[next_index]
+            if not next_char.isalnum():
+                return completion
+
+            text = text[1:] + next_char
+
+            completion += next_char
+            if (len(text + completion) + 2 > len(text) and next_char == ' '):
+                return completion
+
+    def predict(self, search_text):
+        self.unique_chars, self.char_indices, self.indices_char = \
+            DataManager.prepare_corpus(DataManager.get_last_corpus())
+
+        text = search_text[-self.sequence_length:].lower()
+        unique_chars_text = sorted(list(set(text)))
+        for char in unique_chars_text:
+            if char not in self.unique_chars:
+                return []
+
+        if self.sequence_length > len(text):
+            return []
+
+        if not os.path.isfile(self.weights_path):
+            self.logger.debug(
+                'Model weights %s does not exist!' % self.weights_path)
+            return []
+        if not os.path.isfile(self.network_path):
+            self.logger.debug(
+                'Model network does not exist!' % self.network_path)
+            return []
+
+        try:
+            lock_model_network = FileLock('%s.lock' % self.weights_path)
+            lock_model_weigths = FileLock('%s.lock' % self.network_path)
+
+            with lock_model_network.acquire(timeout=1000):
+                with lock_model_weigths.acquire(timeout=1000):
+                    graph = tf.Graph()
+                    with graph.as_default():
+                        session = tf.Session()
+                        with session.as_default():
+                            with open(self.network_path, 'r') as json_file:
+                                loaded_model_json = json_file.read()
+
+                            self.model = model_from_json(loaded_model_json)
+                            self.model.load_weights(self.weights_path)
+
+                            x = self.prepare_input(text)
+                            preds = self.model.predict(x, verbose=0)[0]
+                            next_indices = self.sample(
+                                preds, self.number_predictions)
+
+                            return [
+                                self.indices_char[idx] +
+                                self.predict_completion(
+                                    text[1:] + self.indices_char[idx]
+                                ) for idx in next_indices]
+        except Exception as e:
+            self.logger.debug('Error while prediction: %s' % str(e))
+            return []
diff --git a/ckanext/knowledgehub/lib/rnn/worker.py b/ckanext/knowledgehub/lib/rnn/worker.py
new file mode 100644
index 00000000..0bb548a7
--- /dev/null
+++ b/ckanext/knowledgehub/lib/rnn/worker.py
@@ -0,0 +1,208 @@
+import logging
+import os
+import pickle
+import time
+
+import numpy as np
+import tensorflow as tf
+from filelock import FileLock
+
+from keras.models import Sequential
+from keras.layers import LSTM, Activation
+from keras.layers.core import Dense, Activation
+from keras.optimizers import RMSprop
+from keras.callbacks import EarlyStopping, ModelCheckpoint, ReduceLROnPlateau
+
+from ckanext.knowledgehub.lib.rnn.data_manager import DataManager
+from ckanext.knowledgehub.lib.rnn.config import PredictiveSearchConfig
+
+from ckan.lib.navl import dictization_functions as df
+
+
+np.random.seed(42)
+tf.set_random_seed(42)
+
+
+class PredictiveSearchWorker(PredictiveSearchConfig):
+    ''' A worker that gets the kwh data, create corpus(training data),
+    prepare the data for training, traing the model and save it.
+    '''
+
+    def __init__(self):
+        super(PredictiveSearchWorker, self).__init__()
+        self.model = Sequential()
+        self.history = None
+        self.corpus = None
+        self.training_data = None
+        self.unique_chars = None
+        self.char_indices = None
+        self.x = None
+        self.y = None
+        self.logger = logging.getLogger('ckanext.PredictiveSearchWorker')
+
+    def __check_if_paths_exist(self):
+        weights_dir = os.path.dirname(self.weights_path)
+        if not os.path.exists(weights_dir):
+            os.makedirs(weights_dir)
+        netwokr_dir = os.path.dirname(self.network_path)
+        if not os.path.exists(netwokr_dir):
+            os.makedirs(netwokr_dir)
+        history_dir = os.path.dirname(self.history_path)
+        if not os.path.exists(history_dir):
+            os.makedirs(history_dir)
+
+        return self
+
+    def __set_corpus(self):
+        self.corpus = DataManager.get_corpus()
+        self.training_data = self.corpus.lower()
+        self.training_data = ''.join(
+            ch for ch in self.training_data if ch.isalnum() or ch == ' ')
+        return self
+
+    def __validate_corpus(self):
+        if self.corpus_length > len(self.training_data):
+            msg = ('The minimum length of the corpus is {}, '
+                   'current corpus has {}').format(
+                       int(self.corpus_length), len(self.training_data))
+            self.logger.debug(msg)
+            raise df.Invalid(msg)
+
+        if len(self.training_data) <= self.sequence_length:
+            msg = (
+                'The length of the RNN sequence %d, given %d'
+                % (int(sequence_length), len(self.training_data))
+            )
+            self.logger.debug(msg)
+            raise df.Invalid(msg)
+
+        return self
+
+    def __process_corpus(self):
+        self.unique_chars, self.char_indices, _ = \
+            DataManager.prepare_corpus(self.training_data)
+        return self
+
+    def __set_x_y(self):
+        sentences = []
+        next_chars = []
+        for i in range(
+                0, len(self.training_data) - self.sequence_length, self.step):
+            sentences.append(self.training_data[i: i + self.sequence_length])
+            next_chars.append(self.training_data[i + self.sequence_length])
+
+        self.logger.info('Number of training examples: %d ' % len(sentences))
+
+        self.x = np.zeros((
+            len(sentences),
+            self.sequence_length,
+            len(self.unique_chars)),
+            dtype=np.bool
+        )
+        self.y = np.zeros(
+            (len(sentences), len(self.unique_chars)), dtype=np.bool)
+        for i, sentence in enumerate(sentences):
+            for t, char in enumerate(sentence):
+                self.x[i, t, self.char_indices[char]] = 1
+            self.y[i, self.char_indices[next_chars[i]]] = 1
+        return self
+
+    def __prepare_model(self):
+        self.model.add(LSTM(128, input_shape=(
+            self.sequence_length, len(self.unique_chars))))
+        self.model.add(Dense(len(self.unique_chars)))
+        self.model.add(Activation('softmax'))
+
+        optimizer = RMSprop(lr=0.01)
+        self.model.compile(
+            loss='categorical_crossentropy',
+            optimizer=optimizer,
+            metrics=['accuracy'])
+
+        return self
+
+    def __train_model(self):
+        earlyStopping = EarlyStopping(
+            monitor='val_loss',
+            patience=10,
+            verbose=0,
+            mode='min'
+                )
+        mcp_save = ModelCheckpoint(
+            self.temp_weigths_path,
+            save_best_only=True,
+            save_weights_only=True,
+            monitor='val_loss',
+            mode='min'
+        )
+        reduce_lr_loss = ReduceLROnPlateau(
+            monitor='val_loss',
+            factor=0.1,
+            patience=7,
+            verbose=1,
+            epsilon=1e-4,
+            mode='min'
+        )
+
+        try:
+            self.history = self.model.fit(
+                self.x,
+                self.y,
+                validation_split=0.05,
+                batch_size=128,
+                epochs=self.epochs,
+                shuffle=True,
+                callbacks=[
+                    earlyStopping,
+                    mcp_save,
+                    reduce_lr_loss
+                ]
+            ).history
+        except Exception as e:
+            self.logger.debug('Error while training the model: %s' % str(e))
+            raise e
+
+        return self
+
+    def __save_model(self):
+        try:
+            lock_model_network = FileLock('%s.lock' % self.network_path)
+            lock_model_weigths = FileLock('%s.lock' % self.weights_path)
+
+            with lock_model_network.acquire(timeout=1000):
+                with lock_model_weigths.acquire(timeout=1000):
+                    if os.path.exists(self.weights_path):
+                        os.remove(self.weights_path)
+                    os.rename(self.temp_weigths_path, self.weights_path)
+
+                    with open(self.network_path, "w") as json_file:
+                        model_json = self.model.to_json()
+                        json_file.write(model_json)
+
+                    DataManager.create_corpus(self.training_data)
+        except Exception as e:
+            self.logger.debug('Error while saving RNN model: %s' % str(e))
+            raise e
+        finally:
+            os.remove('%s.lock' % self.network_path)
+            os.remove('%s.lock' % self.weights_path)
+
+        return self
+
+    def __save_history(self):
+        try:
+            pickle.dump(self.history, open(self.history_path, "wb"))
+        except Exception as e:
+            self.logger.debug('Error while saving RNN history: %s' % str(e))
+            raise e
+
+    def run(self):
+        self.__check_if_paths_exist().\
+             __set_corpus().\
+             __validate_corpus().\
+             __process_corpus().\
+             __set_x_y().\
+             __prepare_model().\
+             __train_model().\
+             __save_model().\
+             __save_history()
diff --git a/ckanext/knowledgehub/logic/action/create.py b/ckanext/knowledgehub/logic/action/create.py
index ea63827d..7e8ef875 100644
--- a/ckanext/knowledgehub/logic/action/create.py
+++ b/ckanext/knowledgehub/logic/action/create.py
@@ -101,7 +101,23 @@ def theme_create(context, data_dict):
     session.add(theme)
     session.commit()
 
-    return _table_dictize(theme, context)
+    theme_data = _table_dictize(theme, context)
+
+    # Add to kwh data
+    try:
+        data_dict = {
+            'type': 'theme',
+            'title': theme_data.get('title'),
+            'description': theme_data.get('description'),
+            'theme': theme_data.get('id')
+        }
+        logic.get_action('kwh_data_create')(context, data_dict)
+    except Exception as e:
+        log.debug('Unable to store theme {} to knowledgehub data: {}'.format(
+            theme_data.get('id'), str(e)
+        ))
+
+    return theme_data
 
 
 @toolkit.side_effect_free
@@ -139,7 +155,24 @@ def sub_theme_create(context, data_dict):
     st = SubThemes(**data)
     st.save()
 
-    return st.as_dict()
+    sub_theme_data = st.as_dict()
+
+    # Add to kwh data
+    try:
+        data_dict = {
+            'type': 'sub_theme',
+            'title': sub_theme_data.get('title'),
+            'description': sub_theme_data.get('description'),
+            'sub_theme': sub_theme_data.get('id')
+        }
+        logic.get_action('kwh_data_create')(context, data_dict)
+    except Exception as e:
+        log.debug(
+            'Unable to store sub-theme {} to knowledgehub data: {}'.format(
+                sub_theme_data.get('id'), str(e)
+        ))
+
+    return sub_theme_data
 
 
 def research_question_create(context, data_dict):
@@ -210,6 +243,20 @@ def research_question_create(context, data_dict):
         ResearchQuestion.delete(research_question.id)
         raise e
 
+    # Add to kwh data
+    try:
+        data_dict = {
+            'type': 'research_question',
+            'title': research_question_data.get('title'),
+            'research_question': research_question_data.get('id')
+        }
+        logic.get_action('kwh_data_create')(context, data_dict)
+    except Exception as e:
+        log.debug(
+            'Unable to store research question %s to knowledgehub data: %s'
+                % (research_question_data.get('id'), str(e))
+        )
+
     return research_question_data
 
 
@@ -326,6 +373,21 @@ def resource_view_create(context, data_dict):
     # Add to index
     Visualization.add_to_index(rv_data)
 
+    # Add to kwh data
+    try:
+        data_dict = {
+            'type': 'visualization',
+            'title': rv_data.get('title'),
+            'description': rv_data.get('description'),
+            'visualization': rv_data.get('id')
+        }
+        logic.get_action('kwh_data_create')(context, data_dict)
+    except Exception as e:
+        log.debug(
+            'Unable to store visualization %s to knowledgehub data: %s'
+                % (rv_data.get('id'), str(e))
+        )
+
     # this check is because of the unit tests
     if rv_data.get('__extras'):
         ext = rv_data['__extras']
@@ -402,11 +464,42 @@ def dashboard_create(context, data_dict):
     # Add to index
     Dashboard.add_to_index(dashboard_data)
 
+    # Add to kwh data
+    try:
+        data_dict = {
+            'type': 'dashboard',
+            'title': dashboard_data.get('title'),
+            'description': dashboard_data.get('description'),
+            'dashboard': dashboard_data.get('id')
+        }
+        logic.get_action('kwh_data_create')(context, data_dict)
+    except Exception as e:
+        log.debug(
+            'Unable to store dashboard %s to knowledgehub data: %s'
+                % (dashboard_data.get('id'), str(e))
+        )
+
     return dashboard_data
 
 
 def package_create(context, data_dict):
-    return ckan_package_create(context, data_dict)
+    dataset = ckan_package_create(context, data_dict)
+
+    try:
+        data_dict = {
+            'type': 'dataset',
+            'title': dataset.get('title'),
+            'description': dataset.get('notes'),
+            'dataset': dataset.get('id')
+        }
+        logic.get_action('kwh_data_create')(context, data_dict)
+    except Exception as e:
+        log.debug(
+            'Unable to store dataset %s to knowledgehub data: %s'
+                % (dataset.get('id'), str(e))
+        )
+
+    return dataset
 
 
 def resource_feedback(context, data_dict):
@@ -519,13 +612,43 @@ def resource_validation_create(context, data_dict):
         return _table_dictize(rv, context)
 
 
+@toolkit.side_effect_free
 def kwh_data_create(context, data_dict):
+    ''' Store Knowledge Hub data needed for predictove search.
+    It keeps the title and description of KWH entities: themes, sub-themes, research
+    questions, datasets, visualizations and dashboards.
+
+    :param type: the type of the entity, can be:
+    [
+        'search_query',
+        'theme',
+        'sub_theme',
+        'research_question',
+        'dataset',
+        'visualization',
+        'dashboard'
+    ]
+    :type type: string
+    :param title: the title of the entity
+    :type title: string
+    :param description: the description of the entity (optional)
+    :type description: string
+    :param theme: the ID of th theme (optional)
+    :type theme: string
+    :param sub_theme: the ID of the sub theme (optional)
+    :type sub_theme: string
+    :param research_question: the ID of the research question (optional)
+    :type research_question: string
+    :param dataset: the ID of the dataset (optional)
+    :type dataset: string
+    :param visualization: the ID of the visualization (optional)
+    :type visualization: string
+    :param dashboard: the ID of the dashboard (optional)
+    :type dashboard: string
+
+    returns: the newly created data
+    :rtype: dict
     '''
-    Store Knowledge Hub data
-        :param type
-        :param content
-    '''
-    check_access('kwh_data', context, data_dict)
 
     session = context['session']
 
@@ -537,20 +660,21 @@ def kwh_data_create(context, data_dict):
         raise ValidationError(errors)
 
     user = context.get('user')
-    user_data = model.User.by_name(user.decode('utf8'))
-    if user_data:
-        data['user'] = user_data.id
+    if user:
+        data['user'] = model.User.by_name(user.decode('utf8')).id
 
     kwh_data = KWHData.get(
         user=data.get('user'),
-        content=data.get('content'),
-        type=data.get('type')
+        type=data.get('type'),
+        title=data.get('title'),
+        description=data.get('description'),
     ).first()
 
     if not kwh_data:
         kwh_data = KWHData(**data)
         kwh_data.save()
-        return kwh_data.as_dict()
+
+    return kwh_data.as_dict()
 
 
 def corpus_create(context, data_dict):
diff --git a/ckanext/knowledgehub/logic/action/delete.py b/ckanext/knowledgehub/logic/action/delete.py
index bc37b330..734c8cd2 100644
--- a/ckanext/knowledgehub/logic/action/delete.py
+++ b/ckanext/knowledgehub/logic/action/delete.py
@@ -9,18 +9,21 @@
 from ckan.plugins import toolkit
 from ckan.common import _
 from ckan.model import Session
-from ckanext.knowledgehub import helpers as plugin_helpers
+from ckan.logic.action.delete import package_delete as ckan_package_delete
 
 
-from ckanext.knowledgehub.model import Dashboard
-from ckanext.knowledgehub.model import Theme
-from ckanext.knowledgehub.model import SubThemes
-from ckanext.knowledgehub.model import ResearchQuestion
-from ckanext.knowledgehub.model import Visualization
-from ckanext.knowledgehub.model import UserIntents
-from ckanext.knowledgehub.model import ResourceValidate
-from ckanext.knowledgehub.model import Keyword
-from ckanext.knowledgehub import helpers as kwh_helpers
+from ckanext.knowledgehub import helpers as plugin_helpers
+from ckanext.knowledgehub.model import (
+    Dashboard,
+    Theme,
+    SubThemes,
+    ResearchQuestion,
+    Visualization,
+    UserIntents,
+    ResourceValidate,
+    Keyword,
+    KWHData
+)
 
 
 log = logging.getLogger(__name__)
@@ -217,7 +220,7 @@ def member_delete(context, data_dict=None):
         model.repo.commit()
 
     if obj_type == 'package':
-        kwh_helpers.views_dashboards_groups_update(data_dict.get('object'))
+        plugin_helpers.views_dashboards_groups_update(data_dict.get('object'))
 
 
 def keyword_delete(context, data_dict):
@@ -240,3 +243,23 @@ def keyword_delete(context, data_dict):
 
     Session.delete(Keyword.get(keyword['id']))
     Session.commit()
+
+
+def package_delete(context, data_dict):
+    '''Delete a dataset (package).
+
+    This makes the dataset disappear from all web & API views, apart from the
+    trash.
+
+    You must be authorized to delete the dataset.
+
+    :param id: the id or name of the dataset to delete
+    :type id: string
+
+    '''
+
+    ckan_package_delete(context, data_dict)
+    try:
+        KWHData.delete({'dataset': data_dict['id']})
+    except Exception as e:
+        log.debug('Cannot remove dataset from kwh data %s' % str(e))
diff --git a/ckanext/knowledgehub/logic/action/get.py b/ckanext/knowledgehub/logic/action/get.py
index e7ab8ec6..801570d6 100644
--- a/ckanext/knowledgehub/logic/action/get.py
+++ b/ckanext/knowledgehub/logic/action/get.py
@@ -1,4 +1,5 @@
 import logging
+import re
 import os
 import json
 from six import string_types
@@ -29,7 +30,7 @@
     UserProfile,
 )
 from ckanext.knowledgehub import helpers as kh_helpers
-from ckanext.knowledgehub.rnn import helpers as rnn_helpers
+from ckanext.knowledgehub.lib.rnn import PredictiveSearchModel
 from ckanext.knowledgehub.lib.solr import ckan_params_to_solr_args
 from ckan.lib import helpers as h
 from ckan.controllers.admin import get_sysadmins
@@ -714,10 +715,10 @@ def kwh_data_list(context, data_dict):
     rq = data_dict.get('rq', '')
 
     q = data_dict.get('q', '')
-    page_size = int(data_dict.get('pageSize', 1000000))
+    limit = int(data_dict.get('limit', 1000000))
     page = int(data_dict.get('page', 1))
     order_by = data_dict.get('order_by', None)
-    offset = (page - 1) * page_size
+    offset = (page - 1) * limit
 
     kwargs = {}
 
@@ -735,7 +736,7 @@ def kwh_data_list(context, data_dict):
         kwargs['rq'] = rq
 
     kwargs['q'] = q
-    kwargs['limit'] = page_size
+    kwargs['limit'] = limit
     kwargs['offset'] = offset
     kwargs['order_by'] = order_by
 
@@ -743,9 +744,10 @@ def kwh_data_list(context, data_dict):
 
     try:
         db_data = KWHData.get(**kwargs).all()
-    except Exception:
+    except Exception as e:
+        log.debug('Knowledge hub data list: %s' % str(e))
         return {'total': 0, 'page': page,
-                'pageSize': page_size, 'data': []}
+                'limit': limit, 'data': []}
 
     for entry in db_data:
         kwh_data.append(_table_dictize(entry, context))
@@ -753,7 +755,7 @@ def kwh_data_list(context, data_dict):
     total = len(kwh_data)
 
     return {'total': total, 'page': page,
-            'pageSize': page_size, 'data': kwh_data}
+            'limit': limit, 'data': kwh_data}
 
 
 @toolkit.side_effect_free
@@ -770,20 +772,89 @@ def get_last_rnn_corpus(context, data_dict):
         return c.corpus
 
 
+def _get_predictions_from_db(query):
+    number_predictions = int(
+        config.get(
+            u'ckanext.knowledgehub.rnn.number_predictions',
+            3
+        )
+    )
+
+    text = ' '.join(query.split()[-3:])
+    data_dict = {
+        'q': text,
+        'limit': 10,
+        'order_by': 'created_at desc'
+    }
+    kwh_data = toolkit.get_action('kwh_data_list')({}, data_dict)
+
+    def _get_predict(text, query, index):
+        predict = ''
+        if index != -1:
+            index = index + len(query)
+            for i, ch in enumerate(text[index:]):
+                if ch.isalnum() or (i == 0 and ch == ' '):
+                    predict += ch
+                else:
+                    break
+
+        return predict
+
+    def _findall(pattern, text):
+        return [
+            match.start(0) for match in re.finditer(pattern, text)
+        ]
+
+    predictions = []
+    for data in kwh_data.get('data', []):
+        if len(predictions) >= number_predictions:
+            break
+
+        title = data.get('title').lower()
+        for index in _findall(query, title):
+            predict = _get_predict(title, query, index)
+            if predict != '' and predict not in predictions:
+                predictions.append(predict)
+
+        if data.get('description'):
+            description = data.get('description').lower()
+            for index in _findall(query, description):
+                predict = _get_predict(description, query, index)
+                if predict != '' and predict not in predictions:
+                    predictions.append(predict)
+
+    return predictions[:number_predictions]
+
+
 @toolkit.side_effect_free
 def get_predictions(context, data_dict):
-    ''' Returns a list of predictions
-    :param text: the text for which predictions have to be made
-    :type text: string
+    ''' Returns a list of predictions from RNN model and DB based
+    on data store in knowledge hub
+
+    :param query: the search query for which predictions have to be made
+    :type query: string
     :returns: predictions
     :rtype: list
     '''
+    query = data_dict.get('query')
+    if not query:
+        raise ValidationError({'query': _('Missing value')})
+    if len(query) < int(config.get(
+            u'ckanext.knowledgehub.rnn.sequence_length', 10)):
+        return []
 
-    text = data_dict.get('text')
-    if not text:
-        raise ValidationError({'text': _('Missing value')})
+    if query.isspace():
+        return []
+    query = query.lower()
+
+    predictions = _get_predictions_from_db(query)
+
+    model = PredictiveSearchModel()
+    for p in model.predict(query):
+        if p not in predictions:
+            predictions.append(p)
 
-    return rnn_helpers.predict_completions(text)
+    return predictions
 
 
 def _search_entity(index, ctx, data_dict):
diff --git a/ckanext/knowledgehub/logic/action/update.py b/ckanext/knowledgehub/logic/action/update.py
index 7a76b662..e5e0b105 100644
--- a/ckanext/knowledgehub/logic/action/update.py
+++ b/ckanext/knowledgehub/logic/action/update.py
@@ -71,6 +71,8 @@ def theme_update(context, data_dict):
 
     # we need the old theme name in the context for name validation
     context['theme'] = theme.name
+    context['title'] = theme.title
+    context['description'] = theme.description
     session = context['session']
     data, errors = _df.validate(data_dict,
                                 knowledgehub_schema.theme_schema(),
@@ -92,7 +94,23 @@ def theme_update(context, data_dict):
     session.add(theme)
     session.commit()
 
-    return _table_dictize(theme, context)
+    theme_data = _table_dictize(theme, context)
+
+    # Update kwh data
+    try:
+        data_dict = {
+            'type': 'theme',
+            'entity_id': theme_data.get('id'),
+            'title': theme_data.get('title'),
+            'description': theme_data.get('description')
+        }
+        logic.get_action(u'kwh_data_update')(context, data_dict)
+    except Exception as e:
+        log.debug('Unable to update theme {} in knowledgehub data: {}'.format(
+            theme_data.get('id'), str(e)
+        ))
+
+    return theme_data
 
 
 @toolkit.side_effect_free
@@ -126,6 +144,8 @@ def sub_theme_update(context, data_dict):
         raise logic.NotFound(_('Sub-Theme was not found.'))
 
     context['sub_theme'] = sub_theme.name
+    context['title'] = sub_theme.title
+    context['description'] = sub_theme.description
     data, errors = _df.validate(data_dict,
                                 knowledgehub_schema.sub_theme_update(),
                                 context)
@@ -139,7 +159,24 @@ def sub_theme_update(context, data_dict):
     filter = {'id': id}
     st = SubThemes.update(filter, data_dict)
 
-    return st.as_dict()
+    sub_theme_data = st.as_dict()
+
+    # Update kwh data
+    try:
+        data_dict = {
+            'type': 'sub_theme',
+            'entity_id': sub_theme_data.get('id'),
+            'title': sub_theme_data.get('title'),
+            'description': sub_theme_data.get('description')
+        }
+        logic.get_action(u'kwh_data_update')(context, data_dict)
+    except Exception as e:
+        log.debug(
+            'Unable to update sub-theme {} in knowledgehub data: {}'.format(
+            sub_theme_data.get('id'), str(e)
+        ))
+
+    return sub_theme_data
 
 
 def research_question_update(context, data_dict):
@@ -200,6 +237,20 @@ def research_question_update(context, data_dict):
     # Update index
     ResearchQuestion.update_index_doc(rq_data)
 
+    # Update kwh data
+    try:
+        data_dict = {
+            'type': 'research_question',
+            'entity_id': rq_data.get('id'),
+            'title': rq_data.get('title')
+        }
+        logic.get_action(u'kwh_data_update')(context, data_dict)
+    except Exception as e:
+        log.debug(
+            'Unable to update sub-theme {} in knowledgehub data: {}'.format(
+                rq_data.get('id'), str(e)
+        ))
+
     return rq_data
 
 
@@ -320,6 +371,21 @@ def resource_view_update(context, data_dict):
     # Update index
     Visualization.update_index_doc(resource_view_data)
 
+    # Update kwh data
+    try:
+        data_dict = {
+            'type': 'visualization',
+            'entity_id': resource_view_data.get('id'),
+            'title': resource_view_data.get('title'),
+            'description': resource_view_data.get('description')
+        }
+        logic.get_action(u'kwh_data_update')(context, data_dict)
+    except Exception as e:
+        log.debug(
+            'Unable to update visualization %s in knowledgehub data: %s'
+                % (resource_view_data.get('id'), str(e))
+            )
+
     # this check is done for the unit tests
     if resource_view_data.get('__extras'):
         ext = resource_view_data['__extras']
@@ -399,6 +465,21 @@ def dashboard_update(context, data_dict):
     # Update index
     Dashboard.update_index_doc(dashboard_data)
 
+    # Update kwh data
+    try:
+        data_dict = {
+            'type': 'dashboard',
+            'entity_id': dashboard_data.get('id'),
+            'title': dashboard_data.get('title'),
+            'description': dashboard_data.get('description')
+        }
+        logic.get_action(u'kwh_data_update')(context, data_dict)
+    except Exception as e:
+        log.debug(
+            'Unable to update dashboard %s in knowledgehub data: %s'
+            % (dashboard_data.get('id'), str(e))
+        )
+
     return dashboard_data
 
 
@@ -408,20 +489,48 @@ def package_update(context, data_dict):
     '''
     result = ckan_package_update(context, data_dict)
     schedule_data_quality_check(result['id'])
+
+    try:
+        data_dict = {
+            'type': 'dataset',
+            'entity_id': result.get('id'),
+            'title': result.get('title'),
+            'description': result.get('notes')
+        }
+        logic.get_action(u'kwh_data_update')(context, data_dict)
+    except Exception as e:
+        log.debug(
+            'Unable to update dataset {} to knowledgehub data: {}'.format(
+                result.get('id'), str(e)
+            ))
+
     return result
 
 
 def kwh_data_update(context, data_dict):
-    '''
-    Store Knowledge Hub data
-        :param type
-        :param old_content
-        :param new_content
+    ''' Update existing knowledgehub data or create new one
+
+    :param type: the type of the entity, can be:
+    [
+        'search_query',
+        'theme',
+        'sub_theme',
+        'research_question',
+        'dataset',
+        'visualization',
+        'dashboard'
+    ]
+    :param title: the title of the entity
+    :type title: string
+    :param description: the description of the entity (optional)
+    :type description: string
+    :param entity_id: the ID of the entity
+    :type entity_id: string
+    :returns: the updated data
+    :rtype: dict
     '''
     check_access('kwh_data', context, data_dict)
 
-    session = context['session']
-
     data, errors = _df.validate(data_dict,
                                 knowledgehub_schema.kwh_data_schema_update(),
                                 context)
@@ -430,24 +539,32 @@ def kwh_data_update(context, data_dict):
         raise ValidationError(errors)
 
     user = context.get('user')
-    data['user'] = model.User.by_name(user.decode('utf8')).id
+    if user:
+        data['user'] = model.User.by_name(user.decode('utf8')).id
 
-    kwh_data = KWHData.get(
-        user=data['user'],
-        content=data['old_content'],
-        type=data['type']
-    ).first()
+    data_filter = {data['type']: data['entity_id']}
+    kwh_data = KWHData.get(**data_filter).first()
 
     if kwh_data:
         update_data = _table_dictize(kwh_data, context)
         update_data.pop('id')
         update_data.pop('created_at')
-        update_data['content'] = data['new_content']
+        update_data['title'] = data.get('title')
+        update_data['description'] = data.get('description')
 
-        filter = {'id': kwh_data.id}
-        data = KWHData.update(filter, update_data)
+        kwh_data = KWHData.update(data_filter, update_data)
+    else:
+        data_dict = {
+            'user': data.get('user'),
+            'type': data.get('type'),
+            'title': data.get('title'),
+            'description': data.get('description'),
+            data['type']: data['entity_id']
+        }
+        kwh_data = KWHData(**data_dict)
+        kwh_data.save()
 
-        return data.as_dict()
+    return kwh_data.as_dict()
 
 
 def user_intent_update(context, data_dict):
@@ -815,15 +932,18 @@ def resource_validation_revert(context, data_dict):
 
 def tag_update(context, data_dict):
     ''' Update the tag name or vocabulary
+
     :param id: id or name of the tag
     :type id: string
     :param name: the name of the tag
     :type name: string
     :param vocabulary_id: the id of the vocabulary (optional)
     :type vocabulary_id: string
+
     :returns: the updated tag
     :rtype: dictionary
     '''
+
     model = context['model']
 
     try:
@@ -831,7 +951,7 @@ def tag_update(context, data_dict):
     except NotAuthorized:
         raise NotAuthorized(_(u'Need to be system '
                               u'administrator to administer'))
-                              
+
     schema = knowledgehub_schema.tag_update_schema()
     data, errors = _df.validate(data_dict, schema, context)
     if errors:
@@ -849,7 +969,7 @@ def tag_update(context, data_dict):
     session = context['session']
     tag.save()
     session.commit()
-    
+
     return _table_dictize(tag, context)
 
 
diff --git a/ckanext/knowledgehub/logic/auth/create.py b/ckanext/knowledgehub/logic/auth/create.py
index 33633208..967e6010 100644
--- a/ckanext/knowledgehub/logic/auth/create.py
+++ b/ckanext/knowledgehub/logic/auth/create.py
@@ -69,8 +69,8 @@ def kwh_data(context, data_dict):
     '''
         Authorization check for storing KWH data
     '''
-    # sysadmins only
-    return {'success': False}
+    # all login users
+    return {'success': True}
 
 
 def corpus_create(context, data_dict):
diff --git a/ckanext/knowledgehub/logic/auth/delete.py b/ckanext/knowledgehub/logic/auth/delete.py
index 90f0a345..2af9401b 100644
--- a/ckanext/knowledgehub/logic/auth/delete.py
+++ b/ckanext/knowledgehub/logic/auth/delete.py
@@ -1,3 +1,5 @@
+from ckan.logic.auth.delete import package_delete as ckan_package_delete
+
 
 def theme_delete(context, data_dict):
     '''
@@ -54,4 +56,12 @@ def keyword_delete(context, data_dict=None):
     '''
         Authorization check for deletion of a keyword. Sysadmin only.
     '''
-    return {'success': False}
\ No newline at end of file
+    return {'success': False}
+
+
+def package_delete(context, data_dict):
+    '''
+    This auth function must be overriden like this,
+    otherwise a recursion error is thrown.
+    '''
+    return ckan_package_delete(context, data_dict)
diff --git a/ckanext/knowledgehub/logic/schema.py b/ckanext/knowledgehub/logic/schema.py
index ad291d12..34069461 100644
--- a/ckanext/knowledgehub/logic/schema.py
+++ b/ckanext/knowledgehub/logic/schema.py
@@ -135,18 +135,23 @@ def resource_validation_schema():
 def kwh_data_schema():
     return {
         'type': [not_empty, validators.kwh_data_type_validator],
-        'content': [not_empty, unicode],
+        'title': [not_empty, unicode],
+        'description': [ignore_missing, unicode],
         'theme': [ignore_missing, unicode],
         'sub_theme': [ignore_missing, unicode],
-        'rq': [ignore_missing, unicode]
+        'research_question': [ignore_missing, unicode],
+        'dataset': [ignore_missing, unicode],
+        'visualization': [ignore_missing, unicode],
+        'dashboard': [ignore_missing, unicode]
     }
 
 
 def kwh_data_schema_update():
     return {
         'type': [not_empty, validators.kwh_data_type_validator],
-        'old_content': [not_empty, unicode],
-        'new_content': [not_empty, unicode]
+        'title': [not_empty, unicode],
+        'description': [ignore_missing, unicode],
+        'entity_id': [not_empty, unicode]
     }
 
 
diff --git a/ckanext/knowledgehub/logic/validators.py b/ckanext/knowledgehub/logic/validators.py
index c48d6ced..4329cbb4 100644
--- a/ckanext/knowledgehub/logic/validators.py
+++ b/ckanext/knowledgehub/logic/validators.py
@@ -177,14 +177,22 @@ def resource_feedbacks_type_validator(key, data, errors, context):
 
 def kwh_data_type_validator(key, data, errors, context):
     clean_data = df.unflatten(data)
-    rf_type = clean_data.get('type')
-
-    rf_types = ['theme', 'sub-theme', 'rq', 'search']
-
-    if rf_type not in rf_types:
+    entity_type = clean_data.get('type')
+
+    entity_types = [
+        'search_query',
+        'theme',
+        'sub_theme',
+        'research_question',
+        'dataset',
+        'visualization',
+        'dashboard'
+        ]
+
+    if entity_type not in entity_types:
         errors[key].append(
             p.toolkit._(
-                'Allowed KWH data types are: %s' % ', '.join(rf_types)
+                'Allowed KWH data types are: %s' % ', '.join(entity_types)
             )
         )
 
diff --git a/ckanext/knowledgehub/model/kwh_data.py b/ckanext/knowledgehub/model/kwh_data.py
index acb99184..921b4896 100644
--- a/ckanext/knowledgehub/model/kwh_data.py
+++ b/ckanext/knowledgehub/model/kwh_data.py
@@ -21,28 +21,25 @@ class KWHData(DomainObject):
 
     @classmethod
     def get(cls, id_or_name=None, **kwargs):
-        q = kwargs.get('q')
-        limit = kwargs.get('limit')
-        offset = kwargs.get('offset')
-        order_by = kwargs.get('order_by')
-
-        kwargs.pop('q', None)
-        kwargs.pop('limit', None)
-        kwargs.pop('offset', None)
-        kwargs.pop('order_by', None)
+        q = kwargs.pop('q', None)
+        limit = kwargs.pop('limit', None)
+        offset = kwargs.pop('offset', None)
+        order_by = kwargs.pop('order_by', None)
 
         query = Session.query(cls).autoflush(False)
         query = query.filter_by(**kwargs)
 
         if id_or_name:
             query = query.filter(
-                or_(cls.id == id_or_name, cls.content == id_or_name)
+                or_(cls.id == id_or_name, cls.title == id_or_name)
             )
 
         if q:
             query = query.filter(
-                or_(cls.content.contains(q),
-                    cls.content.ilike(r"%{}%".format(q)))
+                or_(cls.title.contains(q),
+                    cls.title.ilike(r"%{}%".format(q)),
+                    cls.description.contains(q),
+                    cls.description.ilike(r"%{}%".format(q)))
             )
 
         if order_by:
@@ -71,7 +68,7 @@ def delete(cls, filter):
             Session.delete(obj)
             Session.commit()
         else:
-            raise logic.NotFound
+            raise logic.NotFound('data not found')
 
 
 kwh_data_table = Table(
@@ -87,9 +84,12 @@ def delete(cls, filter):
         types.UnicodeText,
         nullable=False),
     Column(
-        'content',
+        'title',
         types.UnicodeText,
         nullable=False),
+    Column(
+        'description',
+        types.UnicodeText),
     Column(
         'user',
         types.UnicodeText,
@@ -103,9 +103,21 @@ def delete(cls, filter):
         types.UnicodeText,
         ForeignKey('sub_themes.id', ondelete='CASCADE')),
     Column(
-        'rq',
+        'research_question',
         types.UnicodeText,
         ForeignKey('research_question.id', ondelete='CASCADE')),
+    Column(
+        'dataset',
+        types.UnicodeText,
+        ForeignKey('package.id', ondelete='CASCADE')),
+    Column(
+        'visualization',
+        types.UnicodeText,
+        ForeignKey('resource_view.id', ondelete='CASCADE')),
+    Column(
+        'dashboard',
+        types.UnicodeText,
+        ForeignKey('ckanext_knowledgehub_dashboard.id', ondelete='CASCADE')),
     Column(
         'created_at',
         types.DateTime,
diff --git a/ckanext/knowledgehub/model/theme.py b/ckanext/knowledgehub/model/theme.py
index e28c1658..7503e9b3 100644
--- a/ckanext/knowledgehub/model/theme.py
+++ b/ckanext/knowledgehub/model/theme.py
@@ -44,6 +44,7 @@
 
 
 class Theme(DomainObject):
+    doctype = 'theme'
 
     @classmethod
     def get(cls, reference):
@@ -106,6 +107,36 @@ def delete(cls, filter):
         else:
             raise logic.NotFound(_(u'Theme'))
 
+    @classmethod
+    def add_to_kwh_data(cls, ctx, theme):
+        content = theme.get('title')
+        if theme.get('description'):
+            content += ' %s' % theme.get('description')
+
+        data_dict = {
+            'type': cls.doctype,
+            'content': content,
+            'theme': theme.get('id')
+        }
+        logic.get_action('kwh_data_create')(ctx, data_dict)
+
+    @classmethod
+    def update_kwh_data(cls, ctx, theme):
+        old_content = ctx.get('title')
+        if ctx.get('description'):
+            old_content += ' %s' % ctx.get('description')
+
+        new_content = theme.get('title')
+        if theme.get('description'):
+            new_content += ' %s' % theme.get('description')
+
+        data_dict = {
+            'type': cls.doctype,
+            'old_content': old_content,
+            'new_content': new_content
+        }
+        logic.get_action(u'kwh_data_update')(ctx, data_dict)
+
 
 mapper(Theme, theme_table)
 
diff --git a/ckanext/knowledgehub/rnn/__init__.py b/ckanext/knowledgehub/rnn/__init__.py
deleted file mode 100644
index e69de29b..00000000
diff --git a/ckanext/knowledgehub/rnn/helpers.py b/ckanext/knowledgehub/rnn/helpers.py
deleted file mode 100644
index e6395233..00000000
--- a/ckanext/knowledgehub/rnn/helpers.py
+++ /dev/null
@@ -1,138 +0,0 @@
-import logging
-import os
-
-import heapq
-import numpy as np
-import tensorflow as tf
-from keras.models import load_model
-from filelock import FileLock
-
-from ckan.plugins import toolkit
-from ckan import logic
-from ckan.common import config
-
-
-log = logging.getLogger(__name__)
-
-
-def prepare_rnn_corpus(corpus):
-    unique_chars = sorted(list(set(corpus)))
-    char_indices = dict((c, i) for i, c in enumerate(unique_chars))
-    indices_char = dict((i, c) for i, c in enumerate(unique_chars))
-
-    return unique_chars, char_indices, indices_char
-
-
-def prepare_input(text, unique_chars, char_indices):
-    sequence_length = int(
-        config.get(u'ckanext.knowledgehub.rnn.sequence_length', 10)
-    )
-    x = np.zeros((1, sequence_length, len(unique_chars)))
-    for t, char in enumerate(text):
-        x[0, t, char_indices[char]] = 1.
-
-    return x
-
-
-def sample(preds, top_n=3):
-    preds = np.asarray(preds).astype('float64')
-    preds = np.log(preds)
-    exp_preds = np.exp(preds)
-    preds = exp_preds / np.sum(exp_preds)
-
-    return heapq.nlargest(top_n, range(len(preds)), preds.take)
-
-
-def predict_completion(model, text, unique_chars, char_indices, indices_char):
-    original_text = text
-    generated = text
-    completion = ''
-    while True:
-        x = prepare_input(text, unique_chars, char_indices)
-        preds = model.predict(x, verbose=0)[0]
-        next_index = sample(preds, top_n=1)[0]
-        next_char = indices_char[next_index]
-        text = text[1:] + next_char
-        if next_char in ['?', '!', '.', ',', ' ']:
-            return completion
-
-        completion += next_char
-        if (len(original_text + completion) + 2 > len(original_text) and
-                next_char == ' '):
-
-            return completion
-
-
-def predict_completions(text):
-    text = text.lower()
-    try:
-        c = toolkit.get_action('get_last_rnn_corpus')({}, {}).lower()
-        unique_chars, char_indices, indices_char = prepare_rnn_corpus(c)
-        unique_chars_text = sorted(list(set(text)))
-
-        for char in unique_chars_text:
-            if char not in unique_chars:
-                return []
-    except Exception as e:
-        log.debug('Error while preparing the RNN corpus: %s' % str(e))
-        return []
-
-    sequence_length = int(
-        config.get(u'ckanext.knowledgehub.rnn.sequence_length', 10)
-    )
-    if sequence_length > len(text):
-        return []
-
-    text = text[-sequence_length:].lower()
-
-    model_path = config.get(
-        u'ckanext.knowledgehub.rnn.model',
-        './keras_model.h5'
-    )
-    if not os.path.isfile(model_path):
-        log.debug('Error: RNN model does not exist!')
-        return []
-
-    try:
-        lock_model_path = '%s.lock' % model_path
-        lock = FileLock(lock_model_path)
-        with lock.acquire(timeout=1):
-            n = int(
-                config.get(u'ckanext.knowledgehub.rnn.number_prediction', 3)
-            )
-            graph = tf.Graph()
-            with graph.as_default():
-                session = tf.Session()
-                with session.as_default():
-                    model = load_model(model_path)
-                    x = prepare_input(text, unique_chars, char_indices)
-                    preds = model.predict(x, verbose=0)[0]
-                    next_indices = sample(preds, n)
-                    return [
-                        indices_char[idx] +
-                        predict_completion(
-                            model,
-                            text[1:] + indices_char[idx],
-                            unique_chars,
-                            char_indices,
-                            indices_char
-                        ) for idx in next_indices]
-    except Exception as e:
-        log.debug('Error while prediction: %s' % str(e))
-        return []
-
-
-def get_kwh_data():
-    corpus = ''
-    try:
-        kwh_data = toolkit.get_action('kwh_data_list')({}, {})
-    except Exception as e:
-        log.debug('Error while loading KnowledgeHub data: %s' % str(e))
-        return corpus
-
-    if kwh_data.get('total'):
-        data = kwh_data.get('data', [])
-        for entry in data:
-            corpus += ' %s' % entry.get('content')
-
-    return corpus
diff --git a/ckanext/knowledgehub/rnn/worker.py b/ckanext/knowledgehub/rnn/worker.py
deleted file mode 100644
index 95666afb..00000000
--- a/ckanext/knowledgehub/rnn/worker.py
+++ /dev/null
@@ -1,181 +0,0 @@
-import os
-import logging
-import pickle
-import time
-
-import numpy as np
-import tensorflow as tf
-from filelock import FileLock
-
-from keras.models import Sequential, load_model
-from keras.layers import Dense, Activation
-from keras.layers import LSTM, Dropout
-from keras.layers import TimeDistributed
-from keras.layers.core import Dense, Activation, Dropout, RepeatVector
-from keras.optimizers import RMSprop
-from keras.callbacks import EarlyStopping, ModelCheckpoint, ReduceLROnPlateau
-
-from ckan.common import config
-from ckan.plugins import toolkit
-from ckan.controllers.admin import get_sysadmins
-
-from ckanext.knowledgehub.rnn import helpers as rnn_h
-
-
-np.random.seed(42)
-tf.set_random_seed(42)
-log = logging.getLogger(__name__)
-
-
-def learn():
-    try:
-        original_data = rnn_h.get_kwh_data()
-        data = original_data.lower()
-    except Exception as e:
-        log.debug('Error while training the model: %s' % str(e))
-
-    unique_chars, char_indices, indices_char = rnn_h.prepare_rnn_corpus(data)
-    log.info('unique chars: %d' % len(unique_chars))
-
-    min_length_corpus = int(
-        config.get(u'ckanext.knowledgehub.rnn.min_length_corpus', 300)
-    )
-    if min_length_corpus > len(data):
-        log.info(
-            'The minimum length of the corpus is %s, current corpus has %d'
-            % (min_length_corpus, len(data)))
-        return
-
-    sequence_length = int(
-        config.get(u'ckanext.knowledgehub.rnn.sequence_length', 10)
-    )
-    if len(data) <= sequence_length:
-        log.info(
-            'The length of the RNN sequence %d, given %d'
-            % (sequence_length, len(data)))
-        return
-
-    step = int(config.get(u'ckanext.knowledgehub.rnn.sentence_step', 3))
-    sentences = []
-    next_chars = []
-    for i in range(0, len(data) - sequence_length, step):
-        sentences.append(data[i: i + sequence_length])
-        next_chars.append(data[i + sequence_length])
-
-    log.info('num training examples: %d ' % len(sentences))
-
-    X = np.zeros((
-        len(sentences),
-        sequence_length,
-        len(unique_chars)),
-        dtype=np.bool
-    )
-    y = np.zeros((len(sentences), len(unique_chars)), dtype=np.bool)
-    for i, sentence in enumerate(sentences):
-        for t, char in enumerate(sentence):
-            X[i, t, char_indices[char]] = 1
-        y[i, char_indices[next_chars[i]]] = 1
-
-    history = ''
-    try:
-        model = Sequential()
-        model.add(LSTM(128, input_shape=(sequence_length, len(unique_chars))))
-        model.add(Dense(len(unique_chars)))
-        model.add(Activation('softmax'))
-
-        optimizer = RMSprop(lr=0.01)
-        model.compile(
-            loss='categorical_crossentropy',
-            optimizer=optimizer,
-            metrics=['accuracy'])
-
-        model_path = config.get(
-            u'ckanext.knowledgehub.rnn.model',
-            './keras_model.h5'
-        )
-
-        model_dir = os.path.dirname(model_path)
-        train_module_path = os.path.join(
-            model_dir, 'keras_module_%s.h5' % time.time()
-        )
-        if not os.path.exists(model_dir):
-            try:
-                os.makedirs(model_dir)
-            except Exception as e:
-                log.debug('Error while creating RNN model DIR: %s' % str(e))
-                return
-
-        # earlyStopping, mcp_save and reduce_lr_loss are used to avoid
-        # overfitting and keep only best results
-        earlyStopping = EarlyStopping(
-            monitor='val_loss',
-            patience=10,
-            verbose=0,
-            mode='min'
-        )
-        mcp_save = ModelCheckpoint(
-            train_module_path,
-            save_best_only=True,
-            monitor='val_loss',
-            mode='min'
-        )
-        reduce_lr_loss = ReduceLROnPlateau(
-            monitor='val_loss',
-            factor=0.1,
-            patience=7,
-            verbose=1,
-            epsilon=1e-4,
-            mode='min'
-        )
-        history = model.fit(
-            X,
-            y,
-            validation_split=0.05,
-            batch_size=128,
-            epochs=int(config.get(u'ckanext.knowledgehub.rnn.max_epochs', 50)),
-            shuffle=True,
-            callbacks=[
-                earlyStopping,
-                mcp_save,
-                reduce_lr_loss
-            ]
-        ).history
-    except Exception as e:
-        log.debug('Error while creating RNN model: %s' % str(e))
-        return
-
-    try:
-        sysadmin = get_sysadmins()[0].name
-        context = {'user': sysadmin, 'ignore_auth': True}
-        # model.save(model_path)
-        toolkit.get_action('corpus_create')(context, {
-            'corpus': original_data
-        })
-
-        lock_model_path = '%s.lock' % model_path
-        lock = FileLock(lock_model_path)
-        with lock.acquire(timeout=1000):
-            if os.path.exists(model_path):
-                os.remove(model_path)
-            os.rename(train_module_path, model_path)
-    except Exception as e:
-        log.debug('Error while saving RNN model: %s' % str(e))
-        return
-
-    history_path = config.get(
-        u'ckanext.knowledgehub.rnn.history',
-        './history.p'
-    )
-    history_dir = os.path.dirname(history_path)
-    if not os.path.exists(history_dir):
-        try:
-            os.makedirs(history_dir)
-        except Exception as e:
-            log.debug('Error while creating RNN history DIR: %s' % str(e))
-            return
-
-    try:
-        pickle.dump(history, open(history_path, "wb"))
-    except Exception as e:
-        log.debug('Error while saving RNN history: %s' % str(e))
-        return
diff --git a/ckanext/knowledgehub/tests/test_actions.py b/ckanext/knowledgehub/tests/test_actions.py
index 69271890..382c1958 100644
--- a/ckanext/knowledgehub/tests/test_actions.py
+++ b/ckanext/knowledgehub/tests/test_actions.py
@@ -10,6 +10,7 @@
 from ckan.plugins import toolkit
 from ckan import model
 from datetime import datetime
+from ckan.common import config
 
 from ckanext.knowledgehub.model.theme import theme_db_setup
 from ckanext.knowledgehub.model.research_question import setup as rq_db_setup
@@ -43,8 +44,10 @@
     ResourceValidate,
 )
 from ckanext.knowledgehub.model.keyword import extend_tag_table
+from ckanext.knowledgehub.lib.rnn import PredictiveSearchWorker
 from ckanext.knowledgehub.lib.util import monkey_patch
 from ckanext.datastore.logic.action import datastore_create
+
 from pysolr import Results
 
 assert_equals = nose.tools.assert_equals
@@ -70,14 +73,20 @@ def setup(self):
         rnn_corpus_setup()
         resource_validate_setup()
         os.environ["CKAN_INI"] = './test.ini'
+        extend_tag_table()
+        config['ckanext.knowledgehub.rnn.min_length_corpus'] = 100
+
+        if not plugins.plugin_loaded('knowledgehub'):
+            plugins.load('knowledgehub')
         if not plugins.plugin_loaded('datastore'):
             plugins.load('datastore')
         if not plugins.plugin_loaded('datapusher'):
             plugins.load('datapusher')
-        extend_tag_table()
 
     @classmethod
     def teardown_class(self):
+        if not plugins.plugin_loaded('knowledgehub'):
+            plugins.unload('knowledgehub')
         if not plugins.plugin_loaded('datastore'):
             plugins.unload('datastore')
         if not plugins.plugin_loaded('datapusher'):
@@ -278,11 +287,12 @@ def test_kwh_data_create(self):
 
         data_dict = {
             'type': 'theme',
-            'content': 'Refugees in Syria'
+            'title': 'Refugees in Syria',
+            'description': 'Number of refugees in Syria 2020'
         }
         kwh_data = create_actions.kwh_data_create(context, data_dict)
 
-        assert_equals(kwh_data.get('content'), data_dict.get('content'))
+        assert_equals(kwh_data.get('title'), data_dict.get('title'))
 
     def test_corpus_create(self):
         user = factories.Sysadmin()
@@ -769,9 +779,10 @@ def test_kwh_data_list(self):
 
         data_dict = {
             'type': 'theme',
-            'content': 'Refugees in Syria'
+            'title': 'Refugees in Syria',
+            'description': 'Number of refugees in Syria 2020'
         }
-        kwh_data = create_actions.kwh_data_create(context, data_dict)
+        create_actions.kwh_data_create(context, data_dict)
 
         kwh_data_list = get_actions.kwh_data_list(context, {})
 
@@ -790,7 +801,7 @@ def test_get_last_rnn_corpus(self):
         data_dict = {
             'corpus': 'KHW corpus'
         }
-        kwh_corpus = create_actions.corpus_create(context, data_dict)
+        create_actions.corpus_create(context, data_dict)
 
         last_rnn_corpus = get_actions.get_last_rnn_corpus(context, {})
 
@@ -968,11 +979,56 @@ def test_resource_validate_status(self):
 
         resource_validate_show = get_actions.resource_validate_status(
             context, {'id': rv.get('resource')}
-            )
+        )
 
         assert_equals(
             resource_validate_show.get('what'), data_dict.get('what')
+        )
+
+    def test_get_predictions(self):
+        data_dict = {
+            'type': 'theme',
+            'title': 'Returns Resettlement Protection Social',
+            'description': (
+                'Network Displacement Trends Labor Market Social '
+                'Cohesion Civil Documentation Demographics '
+                'Reception/Asylum Conditions Conditions of Return '
+                'What is the residential distribution of refugees in '
+                'COA? What is the change in total population numbers '
+                'before and after the crisis? What is the breakdown '
+                'of refugees by place of origin at governorate level?'
+                ' What are the monthly arrival trends by place of '
+                'origin at governorate level? What is the average '
+                'awaiting period in COA prior to registration? What '
+                'are the demographic characteristics of the population?'
             )
+        }
+        create_actions.kwh_data_create(get_context(), data_dict)
+
+        worker = PredictiveSearchWorker()
+        worker.run()
+
+        data_dict = {
+            'name': 'theme-name',
+            'title': 'Test title',
+            'description': 'Test description'
+        }
+        theme = create_actions.theme_create(get_context(), data_dict)
+
+        data_dict = {
+            'type': 'theme',
+            'title': 'Refugees in Syria',
+            'description': 'Number of refugees in Syria 2020',
+            'theme': theme['id']
+        }
+        create_actions.kwh_data_create(get_context(), data_dict)
+
+        predicts = get_actions.get_predictions(
+            get_context(),
+            {'query': 'Refugees in Sy'}
+        )
+
+        assert(len(predicts) > 3)
 
 
 class TestKWHDeleteActions(ActionsBase):
@@ -1189,6 +1245,12 @@ def test_resource_validate_delete(self):
             'Validation report of the resource is deleted.'
             )
 
+    def test_package_delete(self):
+        dataset = create_dataset()
+        r = delete_actions.package_delete(get_context(), {'id': dataset['id']})
+
+        assert_equals(r, None)
+
 
 class TestKWHUpdateActions(ActionsBase):
 
@@ -1515,16 +1577,26 @@ def test_kwh_data_update(self):
             'session': model.Session
         }
 
+        data_dict = {
+            'name': 'theme-name',
+            'title': 'Test title',
+            'description': 'Test description'
+        }
+        theme = create_actions.theme_create(context, data_dict)
+
         data_dict = {
             'type': 'theme',
-            'content': 'Refugees in Syria'
+            'title': 'Refugees in Syria',
+            'description': 'Number of refugees in Syria 2020',
+            'theme': theme['id']
         }
-        kwh_data = create_actions.kwh_data_create(context, data_dict)
+        create_actions.kwh_data_create(context, data_dict)
 
         data_dict = {
             'type': 'theme',
-            'old_content': 'Refugees in Syria',
-            'new_content': 'Refugees in Syria updated'
+            'entity_id': theme['id'],
+            'title': 'Refugees in Syria',
+            'description': 'Refugees in Syria updated'
         }
 
         kwh_data_updated = update_actions.kwh_data_update(
@@ -1533,8 +1605,8 @@ def test_kwh_data_update(self):
         )
 
         assert_equals(
-            kwh_data_updated.get('content'),
-            data_dict.get('new_content')
+            kwh_data_updated.get('description'),
+            data_dict.get('description')
         )
 
     def test_resource_validate_update(self):
diff --git a/ckanext/knowledgehub/tests/test_rnn.py b/ckanext/knowledgehub/tests/test_rnn.py
index e9f0f09d..c5d11ec4 100644
--- a/ckanext/knowledgehub/tests/test_rnn.py
+++ b/ckanext/knowledgehub/tests/test_rnn.py
@@ -1,14 +1,10 @@
 """Tests for rnn/worker.py."""
 
 import os
-import mock
 import nose.tools
 
-from ckan.tests import factories
 from ckan import plugins
 from ckan.tests import helpers
-from ckan.plugins import toolkit
-from ckan import model
 from ckan.common import config
 
 from ckanext.knowledgehub.model.theme import theme_db_setup
@@ -23,12 +19,9 @@
     setup as kwh_data_setup
 )
 from ckanext.knowledgehub.logic.action import create as create_actions
-from ckanext.knowledgehub.rnn.helpers import predict_completions
-from ckanext.knowledgehub.rnn.worker import learn
-from ckanext.knowledgehub.logic.action import get as get_actions
-from ckanext.knowledgehub.tests.helpers import (User,
-                                                create_dataset,
-                                                mock_pylons)
+from ckanext.knowledgehub.lib.rnn import PredictiveSearchModel
+from ckanext.knowledgehub.lib.rnn import PredictiveSearchWorker
+from ckanext.knowledgehub.tests.helpers import get_context
 
 assert_equals = nose.tools.assert_equals
 assert_raises = nose.tools.assert_raises
@@ -37,62 +30,77 @@
 
 class ActionsBase(helpers.FunctionalTestBase):
     def setup(self):
-        # helpers.reset_db()
-        # theme_db_setup()
-        # sub_theme_db_setup()
-        # rq_db_setup()
-        # dashboard_db_setup()
-        # resource_feedback_setup()
-        # kwh_data_setup()
-        # rnn_corpus_setup()
+        helpers.reset_db()
+        theme_db_setup()
+        sub_theme_db_setup()
+        rq_db_setup()
+        dashboard_db_setup()
+        resource_feedback_setup()
+        kwh_data_setup()
+        rnn_corpus_setup()
         os.environ["CKAN_INI"] = 'subdir/test.ini'
-        learn()
+        config['ckanext.knowledgehub.rnn.min_length_corpus'] = 100
 
         if not plugins.plugin_loaded('knowledgehub'):
             plugins.load('knowledgehub')
 
 
-class TestRNN(ActionsBase):
-
-    def test_learn(self):
-        user = factories.Sysadmin()
-        context = {
-            'user': user.get('name'),
-            'auth_user_obj': User(user.get('id')),
-            'ignore_auth': True,
-            'model': model,
-            'session': model.Session
-        }
+class TestPredictiveSearchWorker(ActionsBase):
 
+    def test_run(self):
         data_dict = {
             'type': 'theme',
-            'content': ('Demo again Returns Resettlement Protection Social '
-                        'Network Displacement Trends Labor Market Social '
-                        'Cohesion Civil Documentation Demographics '
-                        'Reception/Asylum Conditions Conditions of Return '
-                        'What is the residential distribution of refugees in '
-                        'COA? What is the change in total population numbers '
-                        'before and after the crisis? What is the breakdown '
-                        'of refugees by place of origin at governorate level?'
-                        ' What are the monthly arrival trends by place of '
-                        'origin at governorate level? What is the average '
-                        'awaiting period in COA prior to registration? What '
-                        'are the demographic characteristics of the '
-                        'population?')
+            'title': 'Returns Resettlement Protection Social',
+            'description': (
+                'Network Displacement Trends Labor Market Social '
+                'Cohesion Civil Documentation Demographics '
+                'Reception/Asylum Conditions Conditions of Return '
+                'What is the residential distribution of refugees in '
+                'COA? What is the change in total population numbers '
+                'before and after the crisis? What is the breakdown '
+                'of refugees by place of origin at governorate level?'
+                ' What are the monthly arrival trends by place of '
+                'origin at governorate level? What is the average '
+                'awaiting period in COA prior to registration? What '
+                'are the demographic characteristics of the population?'
+            )
         }
-        kwh_data = create_actions.kwh_data_create(context, data_dict)
+        create_actions.kwh_data_create(get_context(), data_dict)
+
+        worker = PredictiveSearchWorker()
+        worker.run()
+
+        assert_equals(os.path.isfile(worker.weights_path), True)
+        assert_equals(os.path.isfile(worker.network_path), True)
 
-        learn()
 
-        model_path = config.get(
-            u'ckanext.knowledgehub.rnn.model',
-            './keras_model.h5'
-        )
+class TestPredictiveSearchModel(ActionsBase):
+
+    def test_predict(self):
+        data_dict = {
+            'type': 'theme',
+            'title': 'Returns Resettlement Protection Social',
+            'description': (
+                'Network Displacement Trends Labor Market Social '
+                'Cohesion Civil Documentation Demographics '
+                'Reception/Asylum Conditions Conditions of Return '
+                'What is the residential distribution of refugees in '
+                'COA? What is the change in total population numbers '
+                'before and after the crisis? What is the breakdown '
+                'of refugees by place of origin at governorate level?'
+                ' What are the monthly arrival trends by place of '
+                'origin at governorate level? What is the average '
+                'awaiting period in COA prior to registration? What '
+                'are the demographic characteristics of the population?'
+            )
+        }
+        create_actions.kwh_data_create(get_context(), data_dict)
 
-        assert_equals(os.path.isfile(model_path), True)
+        worker = PredictiveSearchWorker()
+        worker.run()
 
-    def test_predict_completions(self):
         text = 'Demo again Returns Resettlement'
-        completions = predict_completions(text)
+        model = PredictiveSearchModel()
+        predicts = model.predict(text)
 
-        assert_equals(len(completions), 3)
+        assert_equals(len(predicts), 3)
diff --git a/ckanext/knowledgehub/views/theme.py b/ckanext/knowledgehub/views/theme.py
index 77ba09e9..224178c5 100644
--- a/ckanext/knowledgehub/views/theme.py
+++ b/ckanext/knowledgehub/views/theme.py
@@ -189,18 +189,6 @@ def post(self):
                             errors,
                             error_summary)
 
-        try:
-            kwh_data = {
-                'type': 'theme',
-                'content': theme.get('title'),
-                'theme': theme.get('id')
-            }
-            logic.get_action(u'kwh_data_create')(
-                context, kwh_data
-            )
-        except Exception as e:
-            log.debug('Error while storing KWH data: %s' % str(e))
-
         return h.redirect_to(u'theme.read',
                              name=theme['name'])
 
@@ -218,7 +206,6 @@ def _prepare(self, name):
                 context, data_dict)
             check_access(u'theme_update', context)
             context['id'] = theme['id']
-            context['old_theme'] = theme
         except NotAuthorized:
             return base.abort(403, _(u'Unauthorized'
                                      u' to update theme'))
@@ -273,19 +260,6 @@ def post(self, name):
             return self.get(name, data_dict,
                             errors, error_summary)
 
-        try:
-            old_theme = context.get('old_theme')
-            kwh_data = {
-                'type': 'theme',
-                'old_content': old_theme.get('title'),
-                'new_content': theme.get('title')
-            }
-            logic.get_action(u'kwh_data_update')(
-                context, kwh_data
-            )
-        except Exception as e:
-            log.debug('Error while storing KWH data: %s' % str(e))
-
         return h.redirect_to(u'theme.read', name=theme['name'])
 
 
diff --git a/requirements.txt b/requirements.txt
index c91b5666..3ef6cecf 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -1,6 +1,6 @@
 pymssql==2.1.4
 functools32==3.2.3.post2
-tensorflow==1.13.1
+tensorflow==1.6.0
 keras==2.2.4
 h5py==2.9.0
 filelock==3.0.12