Chat with PDF

A fully local AI-powered chatbot that lets you upload PDF documents and ask questions about their content. Built with RAG (Retrieval Augmented Generation) architecture — runs 100% offline with no API keys required.

Features

• 🔒 100% Local & Private — No data leaves your machine
• 💰 Zero API Costs — Uses Ollama for free local AI
• 📄 Multi-PDF Support — Upload multiple PDFs, search across all
• 🔍 Semantic Search — Finds meaning, not just keywords
• 📑 Source Citations — Shows PDF name and page number for answers
• 🌐 Works Offline — No internet required after setup

---

Architecture

┌─────────────────┐     ┌─────────────────┐     ┌─────────────────┐
│   Next.js App   │────▶│  Express Server │────▶│     Ollama      │
│   (Frontend)    │     │    (Backend)    │     │   (Local LLM)   │
│   Port: 3000    │     │   Port: 8000    │     │   Port: 11434   │
└─────────────────┘     └────────┬────────┘     └─────────────────┘
                                 │
                    ┌────────────┴────────────┐
                    ▼                         ▼
            ┌───────────────┐         ┌───────────────┐
            │    Qdrant     │         │ Redis/Valkey  │
            │ (Vector DB)   │         │ (Job Queue)   │
            │  Port: 6333   │         │  Port: 6379   │
            └───────────────┘         └───────────────┘

---

Tech Stack

Technology	Purpose
Next.js 15	Frontend UI with React
Express.js	Backend REST API
Ollama	Local LLM (llama3.2) & Embeddings (nomic-embed-text)
Qdrant	Vector database for semantic search
Redis/Valkey	Job queue for background processing
BullMQ	Queue management
LangChain	AI framework for RAG pipeline
Docker	Container runtime for databases

---

Prerequisites

• Node.js v18+
• Docker Desktop
• Ollama

---

🚀 Quick Start

1. Clone the repository

git clone https://github.com/yourusername/chat-with-pdf.git
cd chat-with-pdf

2. Install Ollama models

ollama pull llama3.2
ollama pull nomic-embed-text

3. Start Docker containers

cd server
docker-compose up -d

4. Install dependencies

# Server
cd server
pnpm install

# Client
cd ../client
pnpm install

5. Start the application

# Terminal 1 - Server
cd server
pnpm dev

# Terminal 2 - Worker
cd server
pnpm dev:worker

# Terminal 3 - Client
cd client
pnpm dev

6. Open the app

Visit http://localhost:3000

---

📁 Project Structure

chat_with_pdf/
├── client/                 # Next.js frontend
│   ├── app/               # App router pages
│   ├── components/        # React components
│   │   ├── chat.tsx      # Chat interface
│   │   └── File-upload.tsx
│   └── package.json
│
├── server/                 # Express backend
│   ├── index.js           # API endpoints
│   ├── worker.js          # Background job processor
│   ├── docker-compose.yml # Qdrant + Redis
│   ├── uploads/           # Uploaded PDFs
│   └── package.json
│
└── README.md

---

🔄 How It Works

PDF Upload Flow

1. User uploads PDF via the UI
2. Express server saves file and adds job to Redis queue
3. Worker picks up the job and:
   • Extracts text from PDF
   • Splits text into chunks (300 chars each)
   • Creates embeddings using nomic-embed-text
   • Stores embeddings in Qdrant

Chat Flow

1. User asks a question
2. Question is converted to an embedding
3. Qdrant finds the 2 most similar chunks (semantic search)
4. Chunks + question are sent to llama3.2
5. LLM generates answer based on the context
6. Response is returned with source citations

---

🔧 Configuration

Environment Variables

Variable	Default	Description
PORT	8000	Express server port
OLLAMA_BASE_URL	http://localhost:11434	Ollama API URL
QDRANT_URL	http://localhost:6333	Qdrant database URL
REDIS_HOST	localhost	Redis host
REDIS_PORT	6379	Redis port

---

📊 API Endpoints

Method	Endpoint	Description
GET	/	Health check
POST	/upload/pdf	Upload a PDF file
GET	/chat?message=...	Ask a question

---

#Results