📄 PDF Chat Assistant using Node.js + LangChain + Google Gemini

Chat with your PDFs using AI! This open-source project lets you upload any PDF file and ask natural language questions about its content using Google Gemini LLM, LangChain, and Node.js.

Ideal for document analysis, contract review, resume Q&A, and building AI-powered document search systems.

---

⭐ Features

• Upload PDF files and parse content.
• Chunk and embed text using Gemini Embeddings.
• Store embeddings in-memory using LangChain’s MemoryVectorStore.
• Perform similarity search on user questions.
• Query Google Gemini to generate accurate contextual responses.

🚀 Powered by:

Google Gemini, LangChain, FAISS, Express.js, and pdf-parse

📚 Keywords:

PDF Chat Assistant, LangChain Node.js, Gemini LLM, Document Q&A, RAG, AI Chatbot, PDF Parser, Semantic Search

---

🧰 Tech Stack

Component	Technology
Backend	Node.js (Express)
AI/LLM API	Google Gemini (via LangChain)
PDF Parsing	pdf-parse
Embedding Store	LangChain Memory Vector DB
Chunking	Manual with configurable size

---

🛠️ Installation

git clone https://github.com/your-username/pdf-chat-gemini.git
cd pdf-chat-gemini
npm install

🔐 Environment Variables

Create a .env file in the root:

PORT=3000
GEMINI_API_KEY=your_google_gemini_api_key

✅ You need access to Google AI Studio or Vertex AI to get your Gemini API key.

📦 Run the Project

node app.js

📬 API Endpoints

1. Upload PDF

POST /api/chat/upload
Content-Type: multipart/form-data
Form field: pdf (upload a .pdf file)

• Parses PDF
• Extracts text
• Chunks and embeds it into memory

2. Ask a Question

POST /api/chat/ask
Content-Type: application/json

{
  "question": "What is this PDF about?"
}

• Returns answer based on the most relevant chunks from the uploaded PDF

📁 Project Structure

pdf-chat-assistant/
├── app.js
├── .env
├── controllers/
│   └── chatController.js
├── routes/
│   └── chatRoutes.js
├── services/
│   ├── pdfService.js
│   ├── embeddingService.js
│   └── geminiService.js
├── utils/
│   └── chunkText.js
├── uploads/
└── vector-store/

🧠 How It Works

• Upload: PDF gets parsed and split into chunks.
• Embed: Each chunk is turned into a vector using Gemini Embeddings.
• Store: Vectors are stored in a temporary in-memory vector store.
• Query: User asks a question → similar chunks are retrieved.
• Answer: Gemini responds using those chunks as context.

🧪 Example Usage

Use Postman or Insomnia to:

1. Upload a PDF to /api/chat/upload

2. Then POST to /api/chat/ask with your question

Demo Link

Demo

📄 License

MIT License

🙌 Acknowledgements

LangChain

Google Generative AI (Gemini)

pdf-parse