PDF Intelligence — AI-Powered Document Q&A

A production-style Retrieval-Augmented Generation (RAG) application that lets you upload any PDF and ask unlimited questions about it — powered by Google Gemini and Qdrant.

Developed by Najeeb Ullah Khan

---

Features

• Upload once, ask unlimited times — the PDF is parsed and indexed into Qdrant once; every follow-up question only costs one embedding call
• Semantic search — Qdrant vector similarity search retrieves the most relevant chunks from your document
• Document-scoped results — Qdrant payload filters ensure answers only come from your uploaded PDF, never other documents
• Smart chunking — paragraphs are merged into ~600-word semantic chunks for optimal retrieval quality
• Rate-limit protection — exponential backoff retry on Gemini 429 errors; 250 ms throttle between embedding calls
• Production security — Helmet headers, per-route rate limiting, file-type and size validation, zero sensitive data in logs or client responses
• Clean professional UI — teal/emerald design, drag-and-drop upload, copy-to-clipboard responses, responsive layout

---

Architecture

User uploads PDF
      |
      v
POST /upload  -->  pdf-parse  -->  Smart Chunker  -->  Gemini Embedding (x N chunks)
                                                              |
                                                              v
                                                    Qdrant (upsert with documentId)
                                                              |
                                                              v
                                               Returns: { documentId, chunkCount }

User asks a question
      |
      v
POST /ask  -->  Gemini Embedding (x1 question only)
                      |
                      v
             Qdrant search (filtered by documentId, top 5 chunks)
                      |
                      v
             Gemini 2.5 Flash Lite  -->  Returns AI answer

Embedding cost comparison

Approach	80-chunk PDF x 20 questions
Old (re-embed on every question)	1,620 embedding calls
New (embed once, search only)	101 embedding calls

---

Project Structure

Smart PDF ChatBot/
├── server/                      # Express backend
│   ├── index.js                 # Entry point — routes, middleware, startup
│   ├── helpers/
│   │   ├── embedding.js         # Gemini client + retry/backoff
│   │   ├── chunking.js          # Smart text chunker (~600 words/chunk)
│   │   └── qdrant.js            # Qdrant client, storeChunks, searchChunks
│   ├── uploads/                 # Temp PDF storage (auto-deleted after parse)
│   └── package.json
│
└── frontend/                    # React + Vite frontend
    ├── src/
    │   ├── components/
    │   │   ├── AnalyzeCard.jsx  # Two-phase upload + ask UI
    │   │   ├── ResponseCard.jsx # AI answer display with copy button
    │   │   ├── Navbar.jsx
    │   │   ├── Footer.jsx
    │   │   ├── Button.jsx
    │   │   ├── Loader.jsx
    │   │   └── Textarea.jsx
    │   ├── pages/
    │   │   └── Home.jsx
    │   ├── services/
    │   │   └── api.js           # Axios client — uploadPDF() + askQuestion()
    │   └── styles/
    │       └── global.css       # Design tokens, animations
    └── package.json

---

Getting Started

Prerequisites

Tool	Notes
Node.js v18+	https://nodejs.org
npm v9+	Comes with Node.js
Qdrant Cloud account	https://qdrant.tech
Google Gemini API key	https://ai.google.dev

---

1. Clone the repository

git clone https://github.com/Najeeb-Patoana/Smart-PDF-ChatBot.git
cd Smart-PDF-ChatBot

---

2. Create a .env file in the server/ directory (use .env.example as the template):

cp server/.env.example server/.env

GEMINI_API_KEY=your_gemini_api_key_here
QDRANT_URL=https://your-cluster-id.qdrant.io
QDRANT_API_KEY=your_qdrant_api_key_here

# Server port (default: 3000)
PORT=3000

# Comma-separated list of allowed frontend URLs
# Add your production frontend URL here when deploying
CORS_ORIGIN=http://localhost:5173

Important: Never commit .env — it is already listed in .gitignore.
The .env.example file is safe to commit — it contains no real secrets.

Install backend dependencies:

npm install

---

3. Create the Qdrant collection

Start the server first, then open this URL once in your browser:

http://localhost:3000/create-collection

This creates the pdf-docs collection with:

• Vector size: 3072 (Gemini Embedding-2 output dimension)
• Distance metric: Cosine
• Keyword payload index on documentId (required for document-scoped filtering)

You only need to do this once per Qdrant account.

---

4. Install frontend dependencies

cd ../frontend
npm install

---

5. Run the application

Terminal 1 — Backend:

cd server
node index.js

Expected startup output:

[Server] Connecting to Qdrant...
[Qdrant] Collection 'pdf-docs' ready with payload index.
PDF Intelligence server  -->  http://localhost:3000
  POST /upload  -- index a PDF
  POST /ask     -- answer a question

Terminal 2 — Frontend:

cd frontend
npm run dev

Open http://localhost:5173 in your browser.

---

API Reference

POST /upload

Index a PDF document. The document is processed once and stored in Qdrant.

Request: multipart/form-data

Field	Type	Required	Notes
pdf	File	Yes	PDF only, max 10 MB

Success response:

{
  "success": true,
  "message": "Document uploaded and indexed successfully.",
  "documentId": "fd39fa0a-f634-499f-a78f-583ec0736e43",
  "chunkCount": 42
}

---

POST /ask

Ask a question about an already-indexed document. The PDF is not re-uploaded.

Request: application/json

{
  "documentId": "fd39fa0a-f634-499f-a78f-583ec0736e43",
  "question": "What are the main topics covered in this document?"
}

Success response:

{
  "success": true,
  "answer": "Based on the document, the main topics covered are..."
}

---

Error responses

All endpoints return errors in the same shape:

{
  "success": false,
  "message": "Human-readable description of what went wrong."
}

Common error messages:

HTTP Status	Message
400	No PDF file provided / Question cannot be empty / Invalid documentId format
413	File is too large. Maximum size is 10 MB
422	Could not extract text from this PDF (scanned/image-based)
429	API rate limit reached. Please wait 30 seconds and try again
404	No relevant content found for that question
500	An error occurred while processing your request

---

Security

Measure	Details
HTTP security headers	helmet middleware sets CSP, HSTS, X-Frame-Options, X-Content-Type-Options, and more
CORS	Restricted to http://localhost:5173 only — no public API access
Rate limiting	Upload endpoint: 10 requests / 10 minutes. Ask endpoint: 60 requests / minute
File validation	MIME type and extension checked by multer fileFilter; 10 MB hard cap
Input validation	UUID regex on documentId; question length enforced 3–500 characters
Log safety	Logs never contain API keys, full error messages, stack traces, or user data
Error responses	All client-facing errors use safeErrorMessage() — internal details never leak
Secret management	All API keys stored in .env which is gitignored

---

Tech Stack

Layer	Technology
Frontend	React 18, Vite, CSS Modules, Axios, react-icons
Backend	Node.js, Express 5, multer, pdf-parse
AI Embeddings	Google Gemini Embedding-2 (3072 dimensions)
LLM	Google Gemini 2.5 Flash Lite
Vector Database	Qdrant Cloud
Security	Helmet, express-rate-limit

---

License

MIT License — Copyright 2024 Najeeb Ullah Khan