README.md

---
license: mit
title: RAG Document QA
sdk: gradio
emoji: 💻
colorFrom: blue
colorTo: purple
pinned: false
---
# 📄 Document Q&A System with RAG (LLM Pipeline)

🚀 A Retrieval-Augmented Generation (RAG) system that enables natural language question answering over custom documents with source-grounded responses.

---

## 🔥 Demo

### 📂 Upload & Ingestion
![Upload](screenshots/upload.png)

### 💬 Question Answering
![QA](screenshots/qa.png)

### 📚 Retrieved Sources (Grounded Answers)
![Sources](screenshots/sources.png)

---

## 🧠 Overview

This project implements a **Retrieval-Augmented Generation (RAG)** pipeline that allows users to:

- Upload documents (PDF/TXT)
- Ask natural language questions
- Receive context-aware answers
- View source passages used to generate the answer

The system ensures **low hallucination** by grounding responses strictly in retrieved document context.

---

## ⚙️ Architecture

User Query  
↓  
Embedding (Sentence Transformers)  
↓  
FAISS Vector Search (Top-K Retrieval)  
↓  
Re-ranking (CrossEncoder)  
↓  
Context Selection  
↓  
LLM (FLAN-T5)  
↓  
Final Answer + Source Attribution  

---

## ✨ Features

- 📂 Document upload (PDF & TXT)
- 🔍 Semantic search using FAISS
- 🧠 Context-aware answer generation using LLM
- 📊 Re-ranking for improved retrieval quality
- 📚 Source attribution (transparent answers)
- ⚡ Fast inference (<2s latency on CPU)
- 🎯 No hallucination — answers strictly from document

---

## 🛠️ Tech Stack

- **Language:** Python  
- **LLM:** FLAN-T5 (Hugging Face)  
- **Embeddings:** sentence-transformers  
- **Vector DB:** FAISS  
- **Re-ranking:** CrossEncoder (ms-marco)  
- **Framework:** Gradio  
- **Deployment:** Hugging Face Spaces (optional)

---

## 🚀 How It Works

1. **Document Ingestion**
   - Text extraction
   - Cleaning and chunking
   - Embedding generation

2. **Retrieval**
   - Query converted to embedding
   - Top-K relevant chunks retrieved via FAISS

3. **Re-ranking**
   - CrossEncoder improves relevance ordering

4. **Generation**
   - LLM generates answer using retrieved context

5. **Source Attribution**
   - Displays ranked document chunks used

---

## 📦 Installation

```bash
git clone https://github.com/your-username/rag-doc-qa.git
cd rag-doc-qa

python -m venv venv
venv\Scripts\activate   # Windows

pip install -r requirements.txt
```

---

## ▶️ Run Locally

```bash
python app.py
```

Open:
```
http://127.0.0.1:7860
```

---

## 📊 Example Queries

- What is the main topic of the document?
- Summarize the key points.
- What conclusions are drawn?
- Explain machine learning mentioned in the document.

---

## 🎯 Key Highlights

- Combines **retrieval + generation** for accurate answers
- Reduces hallucination via grounded context
- Implements **real-world RAG pipeline used in industry**
- Includes **re-ranking for improved precision**

---

## 📌 Future Improvements

- Multi-document support
- Chat history memory
- Streaming responses
- Hybrid search (BM25 + dense retrieval)

---

## 👨‍💻 Author

**Shubham**

---

## ⭐ If you like this project

Give it a ⭐ on GitHub and feel free to connect!