Mango App API

Idea

The idea is to create a RESTful API and web application for preparing card queries for Kotlin mobile application. Mango App API interacts with Firebase Realtime Database, gets the required data, applies some search and classification algorithms and provides an output JSON for the mobile app. Also, API processes smaller requests such as like someone, change user data etc.

Expected basic functionality

• Database connection and query processing
• Simple requests from Kotlin frontend processing
• Profile search algorithm
• Minimalistic website
  • Login / registration form
  • Beautiful UI/UX
  • Mobile app link

Future expansions

• Better locations
  • GPS timestamp
  • Distance calculation + search matches in a distance range
  • Google Maps API - more location functionality
• Text bio ML classification algorithm
  • Polynomial Bayes / SVM multi-class classification
  • Hidden user classification, improved search
• Database optimization, SQL
• Events nearby
  • Facebook/VK API
  • Local events parsing into JSON
  • Event recommendations
• Message tips to start up a conversation
• Gamification, more mechanics for comfortable UX (experimental, A/B-tests)
• Guidelines violation filter
  • image recognition
  • Keras machine learning classification

Search algorithm main goals

• Provide the best match for a user according to given parameters
• A user gets content even though there is no matches
• All the given profile bunches must be tracked to avoid unwanted card repetitions
• developing

Packages

firebaseConnector.py

Handles all the database interaction, queries processing and data fetching.

Method	Arguments	Description	Returns
requestsProcessor		Main HTTP requests listener, calls other methods to process received requests
submitLike	user_id_send : String, user_id_liked : String	Adds user id into liked user's tree in DB, see database structure (used_id_send into users/{user_id_liked}/liked_me)	Integer (success/fail code)
fetchUsersWithParams	search_params: Params	Fetches user profiles JSONs from DB according to given params	List of User
evaluateUser	users: User, search_params : Params	Calculates how user profile fits required parameters	Float (User weight coefficient)

app.py

Main web application package handling incoming HTTP requests and sending prepared data

test.py

Debug console app for testing HTTP requests on production

Technologies

• Flask - main web framework
• Firebase - noSQL database by Google
• Heroku - web application deploy
• Pyrebase - Firebase python API
• Requests - HTTP requests handling
• Scikit Learn (future) - ML library
• Keras (future) - Python deep learning API

Inspiration

• Amazing Profile Bio ML AI Model
• Distance Between Two Locations
• Deep Learning Web App Using Keras
• Scikit Learn Classifier Calibration
• Amazing Morphological Analysis Library for NLP
• Scikit Learn Support Vector Machine in Details
• Multiclass ML Classification

API requests supported

Request	Arguments	Description	Returns
like	user_id_send, user_id_liked	Like another profile and fill used_id_send into users/{user_id_liked}/liked_me	result_code
get-cards	_user_id_send	Prepares and sends a set of profile cards according to user's search preferences	JSON with a set of users cards

Firebase Realtime Database

Users

Main database node containing all the user profile data

Chats & Messages

Chats and chat messages nodes
* - node header

Heroku deploy

Actual production URL: https://mango-friends.com/

Web application is easy to be deployed in Heroku by using following commands:
$ heroku login
$ git add .
$ git commit -am "Heroku deploy"
$ git push heroku HEAD:master