Website-Chatbot:

Model Flow Chart:

Packages/Tech Used:

• NLTK (Natural Language ToolKit) - Pre-Processing/Extracting Features from Messages data.
• Boto3 - Accessing Datasets on AWS.
• PyTorch - Building Model.
• Docker - Packaging for deployment.
• Torchserve - Model serving.
• Google Cloud Platform - Model deployment and Load balancing.

Project Overview:

Creating a chatbot for recruiters/employers to interact with on my portfolio wesbite. Plan is to create a chatbot that will be trained on typical questions about me: datapoints are questions and the labels are categories like: "Experience", "Hobbies", "Education" etc. When a message is receieved, the model/bot will predict which category the message belongs in and formulate a response based on some pre-determined responses for questions in this category.

This project has been cloned from my Girlfriend chatbot repo hence there will be a large overlap in terms of code. This chatbot will be deployed using Torchserve and GCP.

Step 1: Cloning:

• Cloning my Girlfriend Chatbot

Step 2: Creating Custom Handler:

• Initialize model and load in trained parameters.
• Implement preprocessing, inference, postprocessing functions.
• Add "handle" function to determine entire pipeline from datapoint to prediction.

Step 3: Torchserve:

• Archive model into a .MAR file.
• Start serving the model locally using torchserve.
• Test served model with health check and POST request.

Step 4: Docker:

• Create requirements.txt for model.
• Dockerfile to package model.

Step 5: Fetch API:

• Setup FetchAPI into website Javascript to be able to make requests.