🧠 MediBot v2.0 — Production-Grade Clinical Decision Support System

MediBot is an advanced Retrieval-Augmented Generation (RAG) powered clinical decision support chatbot. It is designed to provide domain-grounded, reliable medical answers by constraining Large Language Models (LLMs) to curated medical literature.

This project emphasizes hallucination reduction, safety, system reliability, and state-of-art UI/UX, making it suitable for both AI/ML Engineer portfolios and clinical research environments.

---

🚀 What's New in v2.0

• Architectural Overhaul: Transitioned from Flask to a high-performance FastAPI asynchronous backend.
• Premium User Interface: A brand new, fully responsive, clinical-grade Light Glassmorphism Theme with interactive components, FontAwesome icons, and typing animations.
• Conversational Memory: Implemented session-based chat history, allowing the AI to understand multi-turn medical inquiries.
• Advanced RAG Pipeline: Intelligent context retrieval filtering (relevance score > 0.3) and precise source citations (document & page mapping) directly in the UI.
• Optimized Data Ingestion: Sentence-aware chunking with MD5 hash deduplication for cleaner vector stores, processed natively through Pinecone.
• Production-Ready Docker: Upgraded Dockerfile with optimized layer caching, system dependencies, and integrated application health checks.

---

🎯 Motivation

Large Language Models demonstrate strong reasoning abilities but are unreliable in high-risk domains such as healthcare, where hallucinated responses can cause serious harm.

This project investigates how retrieval grounding, architectural constraints, and robust system prompts can:

• Improve factual correctness
• Eliminate speculative generation
• Enable safe clinical decision support without model fine-tuning

---

🏗️ System Architecture

High-Level RAG Pipeline

1. Ingestion: Medical PDFs are recursively parsed, cleaned, and split using sentence-boundary-aware chunking.
2. Embedding: Text chunks are batch-encoded into 384-dimensional dense vectors using SentenceTransformers.
3. Storage: Vectors and serializable metadata are upserted into a Pinecone Serverless Vector Database.
4. Retrieval: User queries are embedded, matched via cosine similarity, and strictly filtered.
5. Generation: Groq (LLaMA 3.3 70B) receives the user query, chat history, and context to generate a safely constrained clinical response.

---

🧪 Methodology

Data Preprocessing

• Source: Directory-based PDF extraction.
• Chunking Strategy: Recursive character splitting.
  • Chunk size: 800 characters
  • Overlap: 150 characters
• Quality Control: Built-in text cleaning and node deduplication via MD5 hashing.

Vector Representation

• Model: all-MiniLM-L6-v2
• Metric: Cosine similarity.

RAG Constraints & Safety

• Prompt Engineering: Clinical-grade persona prompt with strict instructions to route emergencies and decline non-medical questions.
• Context Filtering: Only fragments with a similarity score > 0.3 are injected into the prompt.
• Source Transparency: All bot responses natively display the exact source document to the user.

---

🛠️ Tech Stack

Core Technologies

• Language: Python 3.12+
• LLM Engine: Groq API (LLaMA 3.3 70B Versatile)
• Vector Database: Pinecone (Serverless)
• Embeddings: HuggingFace sentence-transformers
• Backend Framework: FastAPI & Uvicorn
• Frontend: Vanilla JS, HTML5, Modern CSS (CSS Variables, Flexbox)

Deployment & MLOps

• Containerization: Docker
• Repository: Docker Hub
• CI/CD: GitHub Actions (Automated building & pushing)

---

⚙️ Installation & Setup

1. Clone the repository

git clone https://github.com/rashedulalbab253/Clinical-Decision-Support-System-RAG-Powered-Medical-Chatbot.git
cd Clinical-Decision-Support-System-RAG-Powered-Medical-Chatbot

2. Create environment

conda create -n medibot python=3.12 -y
conda activate medibot

3. Install requirements

pip install -r requirements.txt

4. Configuration

Create a .env file in the root directory:

PINECONE_API_KEY="your_pinecone_api_key_here"
GROQ_API_KEY="your_groq_api_key_here"

5. Ingest Data

Process your local PDFs and build the Pinecone index:

python store_index.py

6. Run Application

Start the FastAPI server:

uvicorn app:app --host 0.0.0.0 --port 8080 --reload

Open http://localhost:8080 in your web browser.

---

🐳 Docker Support

Build Image Locally

docker build -t medibot .

Run Container

docker run -d -p 8080:8080 --name medibot -e PINECONE_API_KEY="your_api_key" -e GROQ_API_KEY="your_api_key" medibot

---

🚀 CI/CD Pipeline

This repository uses GitHub Actions to automatically build and push the Docker image to Docker Hub on every push to the main branch.

Required GitHub Repository Secrets:

• DOCKER_USERNAME: Your Docker Hub username
• DOCKER_PASSWORD: Your Docker Hub Access Token
• (Optional) Pinecone and Groq keys can be passed as runtime variables.

Image Registry: rashedulalbab1234/medibot:latest

---

Designed & Developed by Rashedul Albab