ChatDocs

It is an AI-powered web application that lets users upload PDF documents and have intelligent conversations with them. Ask questions in plain English, get instant accurate answers with exact page references.

---

The Problem:

• Humans waste millions of hours every day searching through documents for information that takes seconds to find the text.
• Finding specific information inside large documents is slow, painful, and extremely inefficient.

We all deal with large documents every day.

• Student | 500 page textbook | Exam tomorrow. Cannot find specific topic. Reads for hours.
• Employee | 800 page company handbook | Simple policy question. HR is busy. Waits 2 days for reply.
• Researcher | 50 research papers | Needs specific methodology. Reads every paper manually. Takes weeks.
• Lawyer | 2000 page legal contract | Client needs specific clause urgently. CTRL+F misses context.
• Doctor | Medical literature | Needs specific drug interaction info. Cannot search by meaning.

---

The Solution:

• Upload any PDF document.
• Ask any questions.
• Get the exact answer in seconds.
• With the exact page number it came from.

---

Screenshots

• Login Page
• Sidebar Page
• Chat Page
• History Page

---

Tech Stack

Frontend

Technology	Purpose
React.js	User interface
Tailwind CSS	Styling
Axios	API calls
React Router	Navigation

Backend

Technology	Purpose
Node.js	Runtime environment
Express.js	Web framework
MongoDB	Main database
Mongoose	MongoDB
Passport.js	Google OAuth
JWT	Authentication tokens
Multer	File upload handling
pdf-parse	PDF text extraction

AI & Search

Technology	Purpose
Google Gemini API	Text embeddings (gemini-embedding-001)
Groq API	LLM answer generation (llama-3.3-70b-versatile)
Pinecone	Vector database (semantic search)

DevOps

Technology	Purpose
Git + GitHub	Version control
Vercel	Frontend deployment
Render	Backend deployment
MongoDB Atlas	Cloud database

---

Features

Currently Built ✅

• Google OAuth Authentication (Sign in with Google)
• PDF Document Upload (up to 10MB)
• PDF Text Extraction
• RAG Pipeline (Retrieval Augmented Generation)
• Google Gemini Embeddings (gemini-embedding-001)
• Pinecone Vector Database (semantic search)
• Groq AI Answer Generation (llama-3.3-70b-versatile)
• Top-K Retrieval (K=4 most relevant chunks)
• Query Cleaning and Validation
• Context Window Management
• Real-time Streaming Responses (word by word)
• Page References (every answer cites exact page)
• Chat History (persists after logout)
• Multiple Document Management
• Duplicate Document Prevention
• Conversation Delete
• Document Delete (clears vectors too)
• Toast Notifications
• Chat History Awareness (last 6 messages)
• Error Handling (empty PDF, no chunks, irrelevant questions)

---

Architecture:

RAG Pipeline

UPLOAD PHASE (once per document): PDF Upload → Extract Text (pdf-parse) → Split into Chunks (1000 chars, 200 overlap) → Generate Embeddings (Google Gemini) → Store in Pinecone with metadata (page numbers)

QUESTION PHASE (every question):

User Question → Clean Query → Generate Question Embedding (Gemini) → Search Pinecone → Top 4 Similar Chunks → Format Context with Page Numbers → Add Chat History → Stream to Groq AI (llama-3.3-70b-versatile) → Stream Response to Frontend (SSE) → Save to MongoDB

Prerequisites

Node.js v22+
MongoDB Atlas account
Google Cloud Console account
Groq API account (free)
Google AI Studio account (free, for Gemini)
Pinecone account (free)

Environment Variables

PORT=5000
MONGODB_URI=your_mongodb_connection_string
JWT_SECRET=your_super_secret_jwt_key
GOOGLE_CLIENT_ID=your_google_client_id
GOOGLE_CLIENT_SECRET=your_google_client_secret
GOOGLE_CALLBACK_URL=http://localhost:5000/api/auth/google/callback
CLIENT_URL=http://localhost:5173
GROQ_API_KEY=your_groq_api_key
GEMINI_API_KEY=your_gemini_api_key
PINECONE_API_KEY=your_pinecone_api_key
PINECONE_INDEX_NAME=your_pinecone_index_name

Getting Started

# Clone the repository
git clone https://github.com/kaushiki-tripathi/ChatDocs.git

# Go to backend folder
cd ChatDocs/backend

# Install dependencies
npm install

# Create .env file and add your credentials

# Start development server
npm run dev

# Go to frontend folder (in a new terminal)
cd ../frontend

# Install dependencies
npm install

# Start frontend
npm run dev

Author

• Kaushiki Tripathi
• LinkedIn: https://www.linkedin.com/in/kaushikitripathi2005/