Astha Medical Chatbot

Astha is a Flask-based medical question-answering chatbot built with LangChain, Pinecone, OpenAI, and Hugging Face embeddings. It uses a RAG pipeline to retrieve relevant chunks from an indexed medical PDF and generate concise answers through a web chat interface.

Features

• Medical RAG chatbot with Pinecone vector search
• OpenAI chat model for answer generation
• Hugging Face sentence-transformer embeddings
• Responsive chat UI with Flask templates and static CSS
• Dockerized production runtime with Gunicorn
• GitHub Actions CI/CD to Amazon ECR and EC2
• Self-hosted GitHub runner based deployment on EC2

Project Structure

.
├── app.py                         # Flask app and RAG endpoint
├── Dockerfile                     # Production Docker image
├── requirements.txt               # Python dependencies
├── data/
│   └── Medical_book.pdf           # Source document for indexing
├── src/
│   ├── helper.py                  # PDF loading, splitting, embeddings
│   ├── prompt.py                  # RAG prompt template
│   └── store_index.py             # Pinecone indexing script
├── static/
│   └── style.css                  # Chat UI styles
├── templates/
│   └── chat.html                  # Chat page
└── .github/workflows/
    └── cicd.yaml                  # ECR build and EC2 deployment

Tech Stack

• Python 3.12
• Flask 3.1
• LangChain 0.3
• Pinecone
• OpenAI API
• Sentence Transformers
• Docker
• Gunicorn
• AWS ECR + EC2
• GitHub Actions

Environment Variables

Create a .env file locally:

OPENAI_API_KEY=your_openai_api_key
PINECONE_API_KEY=your_pinecone_api_key

For GitHub Actions deployment, configure these repository secrets:

AWS_ACCESS_KEY_ID
AWS_SECRET_ACCESS_KEY
AWS_DEFAULT_REGION
ECR_REPO
OPENAI_API_KEY
PINECONE_API_KEY

ECR_REPO should be only the ECR repository name, not the full URL.

Example:

ECR_REPO=astha-medical-chatbot

Local Setup

Create and activate a virtual environment:

python3.12 -m venv myenv
source myenv/bin/activate

Install dependencies:

pip install -r requirements.txt

Run the Flask app:

python app.py

Open:

http://127.0.0.1:8080

Build the Pinecone Index

Before the chatbot can answer from your PDF, the Pinecone index must be populated.

Run:

python src/store_index.py

The script:

• loads PDFs from data/
• creates text chunks
• embeds chunks using sentence-transformers/all-MiniLM-L6-v2
• uploads vectors to the Pinecone index:

medical-chatbot-index

You only need to rerun this when the source documents or embedding model change.

API Endpoint

The frontend sends chat messages to:

POST /get

Example:

curl -X POST http://127.0.0.1:8080/get \
  -H "Content-Type: application/json" \
  -d '{"message":"What is acne?"}'

Response:

{
  "answer": "Acne is a common skin disease..."
}

Docker

Build the image:

docker build -t astha-medical-chatbot .

Run the container:

docker run -d \
  --name astha-medical-chatbot \
  -e OPENAI_API_KEY="$OPENAI_API_KEY" \
  -e PINECONE_API_KEY="$PINECONE_API_KEY" \
  -p 8080:8080 \
  astha-medical-chatbot

Open:

http://localhost:8080

AWS Deployment

This project uses GitHub Actions with two jobs:

1. Continuous-Integration

   • builds the Docker image
   • pushes it to Amazon ECR

2. Continuous-Deployment

   • runs on the EC2 self-hosted GitHub runner
   • pulls the latest image from ECR
   • stops and removes the old container
   • starts the new container on port 8080

Live app format:

http://YOUR_EC2_PUBLIC_IP:8080

Deployment status note: the previous EC2 instance used for this project was terminated to avoid ongoing AWS charges. To deploy again, create a new EC2 instance, install Docker, configure a new self-hosted GitHub runner, update the security group, and rerun the GitHub Actions workflow.

Make sure the EC2 security group allows:

Custom TCP 8080
SSH 22

Self-Hosted Runner Notes

On EC2, the runner should be installed as a service:

cd ~/actions-runner
sudo ./svc.sh status
sudo ./svc.sh start

Expected status:

active (running)
Connected to GitHub
Listening for Jobs

If it is not installed:

cd ~/actions-runner
sudo ./svc.sh install
sudo ./svc.sh start

Troubleshooting

Waiting for a runner to pick up this job

The EC2 self-hosted runner is offline.

Fix:

cd ~/actions-runner
sudo ./svc.sh start
sudo ./svc.sh status

Numpy is not available

This is usually a NumPy/Torch compatibility issue in Docker. The project pins:

numpy==1.26.4
torch==2.2.2

Rebuild and redeploy the Docker image after dependency changes.

no space left on device

Docker layers filled the EC2 disk.

Fix on EC2:

docker system prune -a -f
docker volume prune -f
df -h

App opens but chat response fails

Check container logs:

docker logs astha-medical-chatbot --tail 100

Also confirm these environment variables are passed into the container:

OPENAI_API_KEY
PINECONE_API_KEY

Production Start Command

The Docker container runs:

gunicorn app:app --workers 1 --threads 2 --timeout 120 --bind 0.0.0.0:8080

Disclaimer

This chatbot is for educational and informational support only. It is not a replacement for professional medical advice, diagnosis, or treatment.