title	emoji	colorFrom	colorTo	sdk	pinned
Medical Rag Chatbot	⚡	red	green	docker	false

🩺 Medical RAG Chatbot

FastAPI · LangChain · Pinecone · Google Gemini

A Retrieval-Augmented Generation (RAG) based medical chatbot that provides grounded, evidence-based medical information using external medical documents stored in Pinecone.

The system is intentionally designed with strict safety constraints:

• Responses are generated only from retrieved medical context
• If no relevant context is found, the chatbot refuses to guess
• The system is for educational purposes only

---

📦 Project Phases

Phase 1: Pinecone Index Setup (Required – One Time)

Before running ingestion or starting the chatbot, a Pinecone index must be created manually via the Pinecone dashboard.

Steps:

1. Log in to the Pinecone web dashboard
2. Create a new index
3. Configure the index with:
   • Embedding dimension matching the embedding model
   • Similarity metric (e.g. cosine)
4. Save the index name

⚠️ Index creation is not handled by this codebase.

---

Phase 2: Data Ingestion (Mandatory – One Time)

All medical knowledge lives in Pinecone.
The chatbot will not work until documents are ingested.

python ingest_pdfs.py

⚠️ If the Pinecone index is empty, the chatbot will intentionally refuse to answer.

---

Phase 3: Vector Database (Pinecone)

• Stores document embeddings
• Performs semantic similarity search
• Acts as the single source of truth

No documents are stored in Docker or FastAPI.

---

Phase 4: Backend API (FastAPI)

Method	Endpoint	Description
GET	/	Load chatbot UI
POST	/get	Stream chatbot responses

---

Phase 5: Query Classification

Queries are classified to decide whether retrieval is required. This reduces hallucinations, cost, and latency.

---

Phase 6: Retrieval-Augmented Generation (RAG)

This is true RAG, not prompt stuffing.

• Context is retrieved dynamically per query
• Injected automatically as {context}
• LLM is restricted to retrieved information only

---

Phase 7: Streaming Responses

Responses are streamed using Server-Sent Events (SSE) for better UX.

---

🔐 Safety & Medical Constraints

• No diagnosis
• No prescriptions
• No guessing without context
• Always includes a medical disclaimer

---

🐳 Docker Usage (Optional)

Docker is optional and used only for runtime. All medical knowledge remains in Pinecone.

---

🔹 Pinecone Namespace Strategy

The project currently operates using a single Pinecone namespace.

Although multi-namespace support is not available right now, the current design makes it easy to introduce this feature later without significant changes to the codebase.

---

⚙️ Environment Variables

PINECONE_API_KEY=your_key
PINECONE_INDEX_NAME=your_index
GEMINI_API_KEY=your_key
PORT=5678

---

🚀 Running the Project

python pinecone_ingession/ingest_pdfs.py         # One-time ingestion
uvicorn app:app --reload

Open: http://localhost:5678

---

⚠️ Disclaimer

This project is for educational purposes only and does not replace professional medical advice.