---
title: Financial_RAG
emoji: 📘
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 6.1.0
app_file: app.py
pinned: true
---

# Financial RAG Question Answering System

This project implements a **Retrieve–Rerank–Generate (RAG)** pipeline for financial question answering using the **FinQA dataset**.

---

## System Overview

The pipeline works in three stages:

1. **Document Retrieval**  
   - Options: **BM25**, **Dense embeddings (SentenceTransformer)**, **Hybrid**  
   - FAISS is used for dense vector search.

2. **Reranking**  
   - Cross-Encoder (**ms-marco-MiniLM-L-6-v2**) reranks the top documents for relevance.

3. **Answer Generation**  
   - Uses **Groq Llama 3.3 (70B)** to generate answers based only on retrieved context.

---

## Features

- Accurate financial question answering  
- Handles numeric and multi-step reasoning  
- Supports hybrid retrieval for best results  
- Lightweight FAISS index for fast semantic search  
- Optional BM25 for exact-match numeric queries

---

## Installation

Install dependencies:

```bash
pip install -r requirements.txt
```

---

## Running Locally

Paste in console
```bash
python app.py
```
Open your browser at the local URL shown in the console (usually http://127.0.0.1:7860).

---

## Usage in the App

1. Enter your **Groq API Key**

2. Type your financial question

3. Choose a retrieval mode: `hybrid`, `bm25`, or `semantic`

4. Press **Ask** to get the answer and view the top retrieved documents