RAG Forge

Enterprise Retrieval-Augmented Generation System

OSC AI/ML Hackathon 2026

Project Overview • Problem Statement • Features • Architecture • Tech Stack • Setup Instructions • Team Details

---

Project Overview

RAG Forge is a production-grade Retrieval-Augmented Generation (RAG) platform that transforms static document libraries into intelligent, searchable knowledge bases.

Unlike conventional document-chat applications, RAG Forge implements a complete enterprise retrieval pipeline consisting of:

• Document Ingestion
• Semantic Chunking
• BM25 Hybrid Retrieval
• Neural Re-Ranking
• Context Construction
• Groq-Powered LLM Inference
• Confidence Scoring
• Source Citations

The platform enables users to upload large document collections, organize them into collections, and retrieve accurate, grounded answers through natural language conversations.

End-to-End Workflow

Document Upload
      ↓
Document Processing
      ↓
Semantic Chunking
      ↓
BM25 Hybrid Retrieval
      ↓
Neural Re-Ranking
      ↓
Context Builder
      ↓
Groq LLaMA 3.3 70B
      ↓
Grounded Answer + Citations + Confidence Score

---

Problem Statement

Organizations store large amounts of valuable information across:

• PDFs
• Technical Documentation
• Reports
• Policies
• Manuals
• Knowledge Bases

Traditional search systems often fail to provide contextual answers and frequently return irrelevant results.

The challenge is to build an intelligent Retrieval-Augmented Generation (RAG) system capable of:

• Handling large document collections
• Retrieving relevant information efficiently
• Re-ranking retrieved content for improved accuracy
• Generating grounded answers with citations
• Providing confidence metrics
• Delivering fast response times

RAG Forge solves these challenges through a complete retrieval and generation architecture optimized for quality, speed, and transparency.

---

Features

Core RAG Pipeline

Component	Description
Document Ingestion	PDF & TXT processing
Semantic Chunking	Context-preserving chunk generation
BM25 Hybrid Retrieval	Keyword + TF-IDF + Proximity scoring
Neural Re-Ranking	Coverage × Density × Phrase scoring
LLM Synthesis	Groq LLaMA 3.3 70B
Source Attribution	Grounded citations
Confidence Scoring	Reliability estimation

Platform Features

• JWT Authentication
• OTP-Based Password Reset
• Collection-Based Knowledge Management
• Query History Tracking
• CSV Export Functionality
• Analytics Dashboard
• Mobile Responsive Interface
• Multi-Collection Support

Hackathon Bonus Features

✅ Hybrid Search

✅ Streaming Responses

✅ Large PDF Support

✅ Neural Re-Ranking

✅ Confidence Scoring

✅ Pipeline Telemetry

✅ Re-Ranker Comparison Mode

✅ Query History Export

✅ Analytics Dashboard

---

Architecture

RAG Pipeline

┌─────────────────────────────────────────────────────────┐
│                      User Query                         │
└───────────────────────┬─────────────────────────────────┘
                        │
                        ▼
┌─────────────────────────────────────────────────────────┐
│                 BM25 Hybrid Retrieval                   │
│  • BM25 Scoring                                         │
│  • TF-IDF Weighting                                     │
│  • Proximity Boosting                                   │
└───────────────────────┬─────────────────────────────────┘
                        │
                        ▼
┌─────────────────────────────────────────────────────────┐
│                  Neural Re-Ranking                      │
│  • Coverage Score                                       │
│  • Density Score                                        │
│  • Phrase Match Score                                   │
└───────────────────────┬─────────────────────────────────┘
                        │
                        ▼
┌─────────────────────────────────────────────────────────┐
│                   Context Builder                       │
└───────────────────────┬─────────────────────────────────┘
                        │
                        ▼
┌─────────────────────────────────────────────────────────┐
│              Groq LLaMA 3.3 70B Inference               │
└───────────────────────┬─────────────────────────────────┘
                        │
                        ▼
┌─────────────────────────────────────────────────────────┐
│      Grounded Answer + Citations + Confidence Score     │
└─────────────────────────────────────────────────────────┘

System Design

┌─────────────────────┐
│     React Frontend  │
└──────────┬──────────┘
           │
           ▼
┌─────────────────────┐
│    FastAPI Backend  │
└──────────┬──────────┘
           │
           ▼
┌─────────────────────┐
│   BM25 Retrieval    │
└──────────┬──────────┘
           │
           ▼
┌─────────────────────┐
│  Neural Re-Ranker   │
└──────────┬──────────┘
           │
           ▼
┌─────────────────────┐
│  Groq LLaMA 3.3 70B │
└──────────┬──────────┘
           │
           ▼
┌─────────────────────┐
│     SQLite DB       │
└─────────────────────┘

---

Tech Stack

Frontend

• React 18
• Vite
• Tailwind CSS
• React Query
• React Router
• Recharts

Backend

• Python 3.11
• FastAPI
• SQLAlchemy
• SQLite
• aiosqlite
• pypdf
• passlib
• bcrypt
• python-jose
• httpx

AI & Retrieval Layer

• BM25 Retrieval Algorithm
• TF-IDF Scoring
• Neural Re-Ranking
• Groq Cloud LPU
• LLaMA 3.3 70B Versatile

Infrastructure

• Replit Deployment
• REST APIs
• JWT Authentication

---

Performance Metrics

Metric	Value
BM25 Retrieval Time	~350ms
Re-Ranking Time	~1ms
LLM Response Time	~700ms
End-to-End Query Time	~1050ms
Confidence Score	48–62%

---

Project Structure

rag-forge-osc/
│
├── ragforge/
│   └── backend/
│       ├── app/
│       │
│       ├── api/
│       │   ├── rag.py
│       │   ├── documents.py
│       │   ├── collections.py
│       │   ├── auth.py
│       │   └── dashboard.py
│       │
│       ├── core/
│       │   ├── config.py
│       │   ├── database.py
│       │   └── security.py
│       │
│       ├── models/
│       │   └── models.py
│       │
│       ├── schemas/
│       │   └── schemas.py
│       │
│       ├── uploads/
│       ├── requirements.txt
│       └── .env.example
│
└── artifacts/
    └── ragforge/
        ├── src/
        │
        ├── pages/
        │   ├── LandingPage.tsx
        │   ├── DashboardPage.tsx
        │   ├── AskPage.tsx
        │   ├── ComparePage.tsx
        │   ├── CollectionsPage.tsx
        │   ├── DocumentsPage.tsx
        │   ├── HistoryPage.tsx
        │   └── SettingsPage.tsx
        │
        ├── components/
        ├── api/
        ├── hooks/
        ├── utils/
        ├── App.tsx
        └── main.tsx

---

Setup Instructions

Prerequisites

• Python 3.11+
• Node.js 20+
• pnpm
• Groq API Key

Clone Repository

git clone https://github.com/spoorti-k-d/rag-forge-osc.git
cd rag-forge-osc

Backend Setup

cd ragforge/backend

pip install -r requirements.txt

cp .env.example .env

Update .env

GROQ_API_KEY=your_api_key
SECRET_KEY=your_secret_key

Run Backend

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

Frontend Setup

pnpm install

Run Frontend

pnpm --filter @workspace/ragforge run dev

Open:

http://localhost:18324

---

Hackathon Evaluation Criteria

Evaluation Criteria	Implementation
Retrieval Quality	BM25 Hybrid Retrieval
Re-Ranking Effectiveness	Neural Re-Ranking
Answer Quality	Groq LLaMA 3.3 70B
Pipeline Observability	Telemetry Dashboard
Production Readiness	Authentication, Collections, Analytics
Hybrid Search	✅
Streaming Responses	✅
Large PDF Support	✅
Confidence Scoring	✅
Re-Ranker Comparison	✅

---

Team Details

Name	Role	Contributions
M. Shankar Reddy	Backend Engineer	Retrieval Engine, Re-Ranking, Groq Integration
G. Mounika	Backend Engineer	Authentication, APIs, Document Processing
D. Spoorti	Frontend Engineer	UI Development, Dashboard, Analytics, Chat Interface

---

Demo Link

Live Application

https://build-checker--cse3592.replit.app/

---

License

MIT License

---

Built with ⚡ for OSC AI/ML Hackathon 2026

Team Eagle Eye