Medical FAQ Bot

Overview

This repository contains the source code for a Medical FAQ Bot, which answers medical questions using OpenAI's GPT-3.5 API and Telegram Bot API. The bot can respond to users' medical queries via both a FastAPI web interface and a Telegram chat interface. The bot is running on an AWS EC2 instance and is currently in its prototype phase, meaning it’s not yet a full MVP but serves as a fair proof of concept for answering medical-related queries.

You can use the bot from this link: Medical FAQ Bot

Tech Stack

• Backend: FastAPI
• Telegram Bot: Python Telegram Bot
• OpenAI GPT-3.5 for answering medical questions
• LangChain: For managing interactions with OpenAI's GPT model
• NLP Preprocessing: NLTK
• Medical Terms Dataset: Custom CSV-based dataset for filtering questions
• Hosting: AWS EC2 with Dockerized application

How It Works

1. Telegram Bot Integration: Users interact with the bot via Telegram. Upon receiving a question, the bot preprocesses the text using NLTK, identifies whether it is a medical question using a custom medical terms dataset, and either responds with a relevant answer or notifies the user if their query falls outside its scope.

2. FastAPI Interface: The bot also exposes an API endpoint using FastAPI. You can post a medical question to /ask in JSON format, and the bot will respond with an answer using the OpenAI GPT-3.5 engine.

3. Medical Terms Filter: The bot limits its responses to medical questions by checking if the user's query contains any words from a predefined list of medical terms. If no medical terms are found, the bot informs the user that it can only answer medical-related queries. While this approach is functional, it’s not the best practice for determining the context of a query. Advanced Natural Language Processing (NLP) techniques, such as sentiment analysis or topic modeling, could provide more accurate results.

4. LangChain Integration: The bot integrates LangChain, a framework for building applications powered by large language models (LLMs). It enables handling the conversational flow and interacting with GPT-3.5 through the ConversationChain class, allowing the bot to respond fluidly to medical questions.

Features

• FastAPI API to allow HTTP-based querying of medical questions.
• Telegram Bot Integration for real-time querying by users.
• Preprocessing of user input to handle tokenization and stopword removal using NLTK.
• Medical Terms Dataset Filtering to restrict answers to relevant medical queries (Note: This is an early-stage approach and can be further enhanced with better NLP techniques).
• LangChain for conversation management, allowing smoother and more contextually-aware conversations with users.

Setup & Running

Prerequisites

• Python 3.9
• AWS EC2 Instance
• Docker

Steps to Run Locally

1. Clone the Repository:

   git clone https://github.com/M-Moeini/LLM-Medical-FAQ-Bot.git
   cd LLM-Medical-FAQ-Bot
   
   

2. Install Dependencies:

   pip install -r requirements.txt
   
   

3. Set Environment Variables: TELEGRAM_TOKEN: Your Telegram bot token (You can get it from @BotFather in Telegram.) OPENAI_API_KEY: Your OpenAI API key (You can get it from openAI Website.)

4. Run the Application:

   uvicorn app:app --host 0.0.0.0 --port 8001 & python telegram_bot.py
   

5. Access it:

The FastAPI app is accessible at:

http://<your-ec2-public-ip>:8001/ask

The Telegram bot can be interacted with via Telegram if you set it up with @BotFather.

Steps to Run it on AWS EC2

Use the following command to copy your project files to the EC2 instance:

1. Copy the files:

scp -i your-key.pem -r /path/to/your/project ec2-user@your-ec2-public-ip:/home/ec2-user

2. Build and Run Docker Container: Navigate to the project directory on your EC2 instance and run the following commands:

docker build -t medical_bot .
docker run -d -e TELEGRAM_TOKEN=<your_telegram_token> -e OPENAI_API_KEY=<your_openai_key> -p 8001:8001 medical_bot

3. Access it:

The FastAPI app is accessible at:

http://<your-ec2-public-ip>:8001/ask

The Telegram bot can be interacted with via Telegram if you set it up with @BotFather.

Future Improvements

Advanced NLP Techniques

Replacing the simple keyword-based approach for filtering medical queries with more sophisticated NLP techniques such as sentiment analysis, topic modeling, or intent detection.

Full MLOps Pipeline

Currently, this prototype uses a minimal MLOps stack with the help of Docker. Future iterations could include:

• Model versioning to track the performance of different versions of the model.
• Automatic retraining based on feedback or new data.
• Monitoring and alerting to ensure smooth operation of the model in production.

Cloud Development

Running it on AWS ElasticBeans or running it with the help of AWS Fargate + AWS ECS

Contribution

Feel free to fork this repository and contribute by submitting a pull request. For major changes, please open an issue first to discuss what you would like to change.

Contact

• Mahdi Moeini
  • Email: mmoeini@mun.ca
  • LinkedIn: linkedin.com/in/mmoeini
  • GitHub: m-moeini.github.io