🧠 PDF RAG Chatbot with Multi-User Support

A Retrieval-Augmented Generation (RAG) chatbot built with Next.js, enabling multiple users to upload PDF files and query their contents in real-time using AI-powered language models and vector search.
Each user session is securely isolated, ensuring privacy and independent interactions.

---

📦 Features

• 📄 PDF Upload & Processing – Upload and extract text automatically
• 🔍 Smart Document Chunking – Intelligent segmentation for better retrieval
• 🧠 AI-Powered Embeddings – Generate semantic vectors via Hugging Face Transformers
• 💾 Flexible Storage – MongoDB Atlas for vector storage with local fallback
• 🤖 Advanced LLM Integration – Context-aware responses via OpenAI / OpenRouter
• 👥 Multi-User Support – Session-based isolation for concurrent users
• 🔒 Privacy & Security – Each user’s data remains private and independent
• ⚡ Real-Time Chat – Fast, responsive, and interactive interface
• 🧹 Auto-Cleanup – Sessions expire automatically after 24 hours

---

🚀 Getting Started

1️⃣ Clone the Repository

git clone https://github.com/NKomrade/pdf-rag-bot.git
cd pdf-rag-bot

2️⃣ Install Dependencies

npm install

If you face dependency issues:

npm install --legacy-peer-deps

3️⃣ Configure Environment Variables

Create a .env.local file:

cp .env.example .env.local

Then edit .env.local with your credentials:

# === MongoDB Atlas (Primary Storage) ===
MONGODB_URI=mongodb+srv://<username>:<password>@cluster0.mongodb.net/<dbname>?retryWrites=true&w=majority

# === Hugging Face Transformers (Embeddings) ===
HUGGINGFACE_API_KEY=your-huggingface-api-key

# === OpenAI/OpenRouter (LLM Responses) ===
OPENAI_API_KEY=your-openai-or-openrouter-api-key

# === App Configuration ===
NEXT_PUBLIC_SITE_URL=http://localhost:3000

🧠 Tech Stack

Tech	Purpose
Next.js 14	React framework for frontend + API routes
TypeScript	Type-safe development
LangChain	Document loaders and text splitting
Hugging Face	Text embeddings with Transformers API
OpenAI/OpenRouter	Language model inference for responses
MongoDB Atlas	Vector storage + document metadata
TailwindCSS	Modern styling and responsive design
PDF-Parse	PDF text extraction and processing

---

📸 Screenshot

---

📂 Project Structure

.
├── app/
│   ├── layout.tsx                 # App layout and metadata
│   ├── page.tsx                   # Main chat interface
│   ├── test-session/             # Session testing page
│   │   └── page.tsx
│   └── api/
│       ├── upload/               # PDF upload and processing
│       │   └── route.ts
│       ├── query/                # Chat queries and responses
│       │   └── route.ts
│       └── debug/                # System debugging info
│           └── route.ts
├── components/
│   ├── ChatWindow.tsx            # Main chat interface
│   ├── MessageBubble.tsx         # Individual message display
│   └── PdfUploader.tsx           # PDF upload component
├── lib/
│   ├── embeddings.ts             # Hugging Face embeddings
│   ├── llm.ts                    # OpenAI/OpenRouter LLM calls
│   ├── vectorStore.ts            # Vector similarity search
│   └── sessionManager.ts         # Multi-user session handling
├── temp/                         # Temporary file storage
│   └── chunks.json               # Local storage fallback
├── .env.local                    # Environment variables
├── package.json
└── README.md

---

💬 How It Works

🔄 Multi-User Session Flow:

1. User visits → Gets unique session ID (stored in cookies)
2. Upload PDF → Document chunks stored with session isolation
3. Ask questions → Vector search within user's documents only
4. AI responses → Generated using relevant context from user's PDFs
5. Privacy ensured → Each user only sees their own documents

🏗️ RAG Architecture:

PDF Upload → Text Extraction → Chunking → Embeddings → MongoDB Storage
     ↓
User Query → Embedding → Vector Search → Context Retrieval → LLM Response

📱 Example Usage:

1. Upload a PDF file using the uploader
2. Wait for processing completion (chunks + embeddings created)
3. Ask questions like: "What is the main idea of Chapter 2?"
4. Get AI-powered responses based on your document content

---

🛠 To Do

• Add user authentication system
• Implement persistent chat history storage
• Support multiple PDF uploads per session
• Add document management (list, delete, switch)
• Implement conversation memory across sessions
• Add support for other file formats (DOCX, TXT)
• Optimize embedding model for better accuracy
• Add real-time collaboration features

---

🙌 Acknowledgements

• LangChain - Document processing and text splitting
• Hugging Face - Text embeddings and transformers
• OpenAI - Language model inference
• MongoDB Atlas - Vector storage and database
• Next.js - React framework and API routes
• TailwindCSS - Styling and design system

---