Live Demo: cognidoc.netlify.app

CogniDoc is a full-stack, AI-powered web application that transforms static PDF documents into interactive conversational partners. Users can upload a PDF and ask questions in natural language to receive intelligent, context-aware answers generated by a sophisticated AI pipeline.

The application provides a clean, two-step user experience::

1. Before Upload: A clean interface prompts the user to upload a file.
   

2. After Upload & Query: After processing, the app confirms the document is ready and provides an interface for asking questions and receiving AI-generated answers.

• Dynamic PDF Upload: Users can upload any PDF document for analysis.
• AI-Powered Q&A: Leverages a Retrieval-Augmented Generation (RAG) pipeline to provide accurate, context-aware answers.
• Decoupled Full-Stack Architecture: Built with a separate React frontend and a FastAPI backend for scalability and maintainability.
• Persistent Vector Storage: Utilizes Astra DB to store document embeddings for efficient, low-latency semantic searches.
• Real-time Semantic Search: Finds the most relevant parts of the document to answer a user's question, no matter how it’s phrased.

cognidoc/
│
├── 📂 backend/
│ ├── 📄 main.py             # FastAPI application logic
│ ├── 📄 requirements.txt    # Python dependencies for the backend
│ ├── 📄 .env                # Secret keys for local development
│ └── 📄 .gitignore          # Ignores .venv, .env, and pycache
│
├── 📂 frontend/
│ ├── 📂 public/
│ │ └── 📄 index.html        # Main HTML page for the React app
│ │
│ ├── 📂 src/
│ │ ├── 📂 components/
│ │ │ ├── 📄 FileUploader.js # File upload UI component
│ │ │ └── 📄 QueryForm.js    # Question form component
│ │ │
│ │ ├── 📂 services/
│ │ │ └── 📄 api.js          # Handles all API calls to the backend
│ │ │
│ │ ├── 📄 App.js            # Main component orchestrating the UI
│ │ ├── 📄 App.css           # Styles for the App component
│ │ ├── 📄 index.js          # Entry point for the React app
│ │ └── 📄 index.css         # Global styles
│ │
│ ├── 📄 package.json        # Frontend dependencies
│ ├── 📄 package-lock.json   # Locks dependency versions
│ └── 📄 .gitignore          # Ignores node_modules, build files.
│
└── 📄 README.md             # Project documentation

CogniDoc is built around a Retrieval-Augmented Generation (RAG) pipeline, ensuring that the AI’s answers are grounded in the uploaded document.

• The user uploads a PDF via the React frontend.
• The file is sent to the FastAPI backend.
• The backend extracts all text from the PDF.
• The text is split into smaller, manageable chunks using LangChain.
• Each chunk is converted into a numerical representation (vector embedding) using a Hugging Face sentence-transformer model.
• These embeddings are stored in the Astra DB vector store.

• The user asks a question in the React frontend.
• The question is sent to the FastAPI backend.
• The backend performs a semantic search in Astra DB to find the most relevant text chunks.
• These chunks, along with the original question, are passed as context to the Google Gemini API.
• The Gemini model generates a comprehensive, human-like answer based on the provided context.
• The final answer is sent back to the React frontend and displayed to the user.

To run this project locally, set up both the backend and frontend.

• Node.js (v18 or newer)
• Python 3.11 or newer
• Git for cloning the repository
• API keys for Google Gemini and Astra DB

git clone https://github.com/Harsh55g/pdf_query.git
cd cognidoc

# navigate to backend 
cd backend

# Use py -3.11 or python3.11 depending on your system
py -3.11 -m venv .venv
.\.venv\Scripts\activate

# Install the requirements
pip install -r requirements.txt

Configure your secrets by creating a .env

GOOGLE_AI_KEY="your_google_api_key_here"
ASTRA_DB_APPLICATION_TOKEN="your_astra_token_here"
ASTRA_DB_ID="your_astra_db_id_here"

Run the backend server.

uvicorn main:app --reload

# navigate to Frontend
cd frontned

# Install dependencies:
npm install

# run the dev environment
npm start

Thank you for taking the time to read through this project and for using CogniDoc wisely.

If you have any feature updates or improvements in mind, I’d be happy to review your pull requests and collaborate to make the project even better!