RAG Demo App

A minimal Retrieval-Augmented Generation (RAG) system built with Django + Qdrant + Gemini + Cohere.
This app lets users upload documents (PDF/DOCX) or paste text, indexes them into Qdrant with embeddings, and provides a chat-like interface to query across documents with reranking and grounding verification.

---

🚀 Live Demo

Hosted at: Render

• First screen loads without console errors.
• You can upload files, ask questions, and see retrieved chunks, reranked results, and grounded answers with citations.

---

🏗️ Architecture

Flow:

1. Upload Text/File → Extract text → Chunk into 800 tokens with 80 overlap.
2. Embeddings → Generated using Google Gemini (768-dim vectors).
3. Vector DB → Stored in Qdrant with COSINE similarity.
4. Retriever → Top-10 chunks retrieved.
5. Reranker → Cohere Rerank API trims to top-3.
6. LLM Answer → Gemini generates grounded response.
7. Verification → Grounding check to reject hallucinations.
8. UI → Tailwind + minimal JS.

---

⚙️ Configuration

• Chunk size: 800 tokens
• Chunk overlap: 80 tokens
• Retriever: Qdrant, top_k = 10
• Reranker: Cohere, top_k = 3
• LLM provider: Gemini (for embeddings + generation)
• Vector dimensions: 768
• Distance metric: Cosine

---

📦 Setup

1. Clone the repo

git clone https://github.com/<your-username>/<your-repo>.git
cd <your-repo>

2. Install dependencies

pip install -r requirements.txt

3. Environment variables

Copy .env.example to .env and set your keys:

QDRANT_URL=...
QDRANT_API_KEY=...
QDRANT_COLLECTION=rag_demo
GEMINI_API_KEY=...
COHERE_API_KEY=...
DJANGO_SECRET_KEY=...

4. Run locally

python manage.py runserver

Visit http://localhost:8000.

5. Deploy (PythonAnywhere example)

• Upload repo to PythonAnywhere.
• Set env vars under Web > WSGI config.
• Run migrations & reload app.

---

📎 Links

• Live Demo: Python Demo Link
• GitHub Repo: Repo Link
• Resume: Google Drive Link