🤖 PDF RAG Chatbot

An intelligent chatbot that lets you upload any PDF and have a real conversation with it — powered by RAG (Retrieval-Augmented Generation), Pinecone Vector DB, and multiple switchable LLM models via OpenRouter.

Tech Stack NestJS Pinecone OpenRouter

• 📄 PDF Upload — Drag & drop any PDF (up to 20MB)
• 🧠 RAG Pipeline — PDF is chunked, embedded, and stored in Pinecone vector database
• 💬 Real-time Streaming — Answers stream token by token like ChatGPT
• 🔄 Dynamic Model Switching — Choose between multiple free LLM models (Nemotron, Gemma, GPT-OSS) directly from the chat UI in real-time
• 🎯 Context-Aware — Answers strictly based on uploaded document
• 🚫 Out-of-Scope Detection — Politely rejects questions unrelated to the PDF
• 📱 Fully Responsive — Works on mobile, tablet, and desktop
• 🌙 Premium Dark UI — Clean, modern interface built with Tailwind CSS

User uploads PDF
 ↓
NestJS Backend receives file
 ↓
pdf-parse extracts text
 ↓
Text split into chunks (1000 tokens, 200 overlap)
 ↓
multilingual-e5-large embeds each chunk (1024-dim)
 ↓
Pinecone stores all vectors (unique namespace per PDF)
 ↓
User asks a question
 ↓
Question embedded → Pinecone similarity search (Top 4 chunks)
 ↓
Context + Question sent to selected LLM via OpenRouter
 ↓
Answer streams back to frontend in real-time (SSE)

pdf-rag-chatbot/
├── frontend/ # Next.js App
│ ├── app/
│ │ ├── page.tsx # Home "/"
│ │ ├── upload/page.tsx # Upload PDF "/upload"
│ │ └── chat/[namespace]/ # Chat "/chat/:id"
│ ├── components/
│ │ ├── chat/ # ChatWindow, MessageBubble, Input
│ │ ├── pdf/ # PDFUploader
│ │ └── ui/ # Sidebar
│ └── hooks/ # useChat, usePDFUpload
│
└── backend/ # NestJS API
 └── src/
 ├── modules/
 │ ├── pdf/ # PDF upload & processing
 │ └── chat/ # RAG query & SSE streaming
 └── services/
 ├── langchain/ # RAG pipeline & embeddings
 ├── pinecone/ # Vector DB operations
 └── gemini/ # LLM integration

• Node.js 18+
• Pinecone account (free tier)
• OpenRouter account (free tier)

git clone https://github.com/razazaheer12/pdf-rag-chatbot.git
cd pdf-rag-chatbot

cd backend
npm install

Create .env file in backend/ folder:

PORT=5000
GEMINI_API_KEY=your_openrouter_api_key
PINECONE_API_KEY=your_pinecone_api_key
PINECONE_INDEX_NAME=pdf-rag-index-v2
PINECONE_DIMENSION=1024
NODE_ENV=development

npm run start:dev

cd frontend
npm install

Create .env.local file in frontend/ folder:

NEXT_PUBLIC_API_URL=http://localhost:5000

npm run dev

http://localhost:3000

1. Upload — User uploads a PDF on /upload page
2. Processing — Backend parses, chunks, and embeds the PDF into Pinecone
3. Chat — User is redirected to /chat/[namespace] page
4. Model Selection — User picks an LLM model from the dropdown (switchable anytime)
5. Query — User asks a question; backend retrieves top 4 relevant chunks
6. Answer — Selected LLM generates a context-aware answer, streamed in real-time

🏠 Home Page

📤 Upload Page

💬 Chat Page

• Chunk Size: 1000 characters with 200 overlap
• Embedding Model: multilingual-e5-large (1024 dimensions)
• Similarity Search: Top-4 chunks retrieved per query
• Similarity Threshold: Score > 0.4 filtered
• LLM Models (switchable via UI):
  • nvidia/nemotron-3-nano-30b-a3b:free (Fast)
  • google/gemma-4-31b-it:free
  • openai/gpt-oss-20b:free
  • nvidia/nemotron-3-super-120b-a12b:free (Large)
  • nvidia/nemotron-3-ultra-550b-a55b:free (Most Powerful)
• Streaming: Server-Sent Events (SSE) for real-time token streaming

Raza Zaheer

• 🌐 Portfolio: raza-zaheer-portfolio-web-developer.vercel.app
• 💼 GitHub: @razazaheer12

This project is open source and available under the MIT License.

⭐ If you found this project helpful, please give it a star on GitHub!