DocuChat — Offline Document Intelligence Platform

A privacy-focused AI system that enables semantic search and question answering over PDF documents using a locally deployed RAG architecture. Built with FastAPI · FAISS · Local LLM · React · Tailwind CSS

⭐ If you find this project useful, consider giving it a star on GitHub!

🏗️ Architecture

┌─────────────────────────────────────────────────────────────────┐
│                        DocuChat Platform                        │
│                                                                 │
│  ┌────────────┐    ┌──────────────┐    ┌────────────────────┐  │
│  │  React +   │    │   FastAPI    │    │   Local AI Engine  │  │
│  │  Vite UI   │───▶│   Backend    │───▶│                   │  │
│  │            │    │              │    │  ┌─────────────┐   │  │
│  │  - Upload  │    │  - /upload   │    │  │   LLM Chat  │   │  │
│  │  - Chat    │    │  - /query    │    │  └─────────────┘   │  │
│  │  - Status  │    │  - /status   │    │  ┌─────────────┐   │  │
│  └────────────┘    │  - /clear    │    │  │ Embeddings  │   │  │
│                    └──────┬───────┘    │  └─────────────┘   │  │
│                           │            └────────────────────┘  │
│                    ┌──────▼───────┐                             │
│                    │    FAISS     │                             │
│                    │ Vector Store │                             │
│                    │ (In-Memory)  │                             │
│                    └─────────────┘                             │
└─────────────────────────────────────────────────────────────────┘

RAG Pipeline Flow

PDF Upload
    │
    ▼
Text Extraction (pypdf)
    │
    ▼
Smart Chunking (600 chars, 100 overlap)
    │
    ▼
Embedding Generation (SentenceTransformers MiniLM)
    │
    ▼
FAISS Index (Cosine similarity search)
    │
   [Ready for queries]
    │
User Question
    │
    ▼
Question → Embedding
    │
    ▼
FAISS Top-K Search (k=5)
    │
    ▼
Context Assembly
    │
    ▼
Strict Prompt → Local LLM
    │
    ▼
Answer (strictly from document context)

---

📁 Project Structure

docuchat/
├── backend/
│   ├── app/
│   │   ├── __init__.py
│   │   ├── main.py          # FastAPI app, routes, CORS
│   │   ├── embeddings.py    # Ollama nomic-embed-text wrapper
│   │   ├── vector_store.py  # FAISS index management
│   │   └── rag.py           # PDF extraction, chunking, LLM query
│   ├── uploads/             # Temporary PDF storage (auto-created)
│   └── requirements.txt
│
├── frontend/
│   ├── src/
│   │   ├── components/
│   │   │   ├── LandingPage.jsx   # Animated hero + features + arch diagram
│   │   │   ├── Sidebar.jsx       # Status, doc info, stats
│   │   │   ├── UploadPanel.jsx   # Drag-drop upload with progress
│   │   │   ├── ChatArea.jsx      # Full chat UI with input
│   │   │   └── MessageBubble.jsx # User/AI message with source citations
│   │   ├── hooks/
│   │   │   └── useStatus.js      # Polls /status endpoint
│   │   ├── utils/
│   │   │   └── api.js            # Axios API calls
│   │   ├── App.jsx               # Root component + state machine
│   │   ├── main.jsx
│   │   └── index.css             # Tailwind + custom styles
│   ├── index.html
│   ├── vite.config.js
│   ├── tailwind.config.js
│   └── package.json
│
└── README.md

---

🚀 Prerequisites

• Python 3.11+
• Node.js 18+ and npm

---

⚙️ Setup Instructions

Step 1 — Clone the Repository

git clone https://github.com/Asifa007/docuchat-rag.git
cd docuchat-rag

Step 2 — Set Up Backend

cd backend

# Create virtual environment
python -m venv venv

# Activate it
source venv/bin/activate       # Linux/macOS
# OR
.\venv\Scripts\activate        # Windows

# Install dependencies
pip install -r requirements.txt

Step 3 — Run Backend

uvicorn app.main:app --reload --host 0.0.0.0 --port 8000

Backend API docs available at:

http://localhost:8000/docs

Step 4 — Set Up Frontend

cd ../frontend

# Install dependencies
npm install

Step 5 — Run Frontend

npm run dev

Frontend runs on:

http://localhost:5173

🌐 Open the App

Visit:

http://localhost:5173

Ensure backend is running on port 8000

Click "Start Chatting"

Upload a PDF (drag-and-drop supported)

Wait for document indexing

Ask questions — responses are generated strictly from retrieved document context

🧪 Testing the API

# Health check
curl http://localhost:8000/

# Check system status
curl http://localhost:8000/status

# Upload a PDF
curl -X POST http://localhost:8000/upload \
  -F "file=@/path/to/your/document.pdf"

# Query
curl -X POST http://localhost:8000/query \
  -H "Content-Type: application/json" \
  -d '{"question": "What is this document about?"}'

# Clear session
curl -X DELETE http://localhost:8000/clear

🌍 Expose with ngrok (Optional)

ngrok http 5173
ngrok http 8000

Update VITE_API_URL in the frontend if exposing backend publicly.

🔧 Configuration

Change Retrieval Parameters

Edit backend/app/rag.py:

CHUNK_SIZE = 600
CHUNK_OVERLAP = 100
TOP_K_CHUNKS = 5
Change LLM Model (If Using Local Model)
LLM_MODEL = "your-local-model-name"
Change Embedding Model
EMBED_MODEL = "your-embedding-model-name"

🎯 Features

Feature	Status
PDF text extraction	✅
Overlapping chunk strategy	✅
Local embedding generation	✅
FAISS cosine similarity search	✅
Grounded (anti-hallucination) prompt system	✅
Fallback response for insufficient context	✅
Drag-and-drop upload	✅
Upload progress bar	✅
Real-time backend status indicator	✅
Typing indicator animation	✅
Source citations per answer	✅
Character counter for input	✅
Mobile-responsive design	✅
Modern glassmorphism UI	✅
Animated landing page	✅
System architecture diagram	✅
Clear & reset session	✅
Auto-scroll chat	✅

---

🔐 Privacy

• ✅ No data sent to any cloud service
• ✅ No internet required after model download
• ✅ PDFs stored locally in backend/uploads/
• ✅ Vectors stored in RAM (cleared on reset)
• ✅ All inference runs on your own hardware

---

📊 Performance Tips

• Smaller PDFs (< 50 pages) typically index in under 30 seconds (CPU-based inference)
• Larger documents (100+ pages) may take 2–5 minutes for embedding generation
• First query may be slower due to model initialization
• Subsequent queries are significantly faster due to cached index
• GPU acceleration (if supported by your local model) improves embedding and inference speed

---

🌐 Live Demo

Frontend
https://docuchat-frontend-h83f.onrender.com

Backend API
https://docuchat-backend-qsf2.onrender.com

🧱 Tech Stack

LLM / Embeddings	SentenceTransformers MiniLM
LLM	Local LLM inference
Embeddings	Local embedding model
Vector DB	FAISS
Backend	FastAPI + Python
PDF Parsing	pypdf
Frontend	React 18 + Vite
Styling	Tailwind CSS
HTTP Client	Axios
Icons	Lucide React
Fonts	Syne + DM Sans + JetBrains Mono

---

🐛 Troubleshooting

Backend not running

uvicorn app.main:app --reload --port 8000

Ensure the backend is accessible at:

http://localhost:8000

• PDF has no extractable text
• The document may be scanned (image-based)
• Use an OCR tool before uploading
• Slow indexing
• Large PDFs generate more chunks and embeddings
• CPU-based systems may take longer
• Reduce CHUNK_SIZE if needed

---

📄 License

MIT License — free to use, modify, and distribute.

---

👩‍💻 Author

Built by Asifa Firdhouse

---