PDFhelper - talk about your PDF!

Live Demo

• Frontend:
  • Next.js (React 18 App Router), Tailwind CSS, React Query
• Backend:
  • FastAPI for API endpoints and cross-origin interaction
• Database:
  • PDFs stored securely in Supabase buckets
  • SQLite database for metadata
  • Local storage for FAISS vector indices
• NLP and Retrieval:
  • LangChain and LlamaIndex for context retrieval
• Embedding Model:
  • togethercomputer/m2-bert-80M-32k-retrieval (Together AI API)
• LLM Model:
  • mistralai/Mixtral-8x7B-Instruct-v0.1 (Together AI API)

Backend Architecture

• FastAPI Endpoints:
  • /upload: Upload PDF documents, save to Supabase Storage, and index locally.
  • /ask: Submit questions and receive streamed answers.
• PDF Processing:
  • Extract text using PyMuPDF.
  • Store temporary vectors using FAISS.
• NLP Handling:
  • LangChain and LlamaIndex manage context retrieval and LLM interaction.

Frontend Architecture

• Next.js App Router based pages and layouts
• React Query for robust asynchronous API handling
• Tailwind CSS for responsive and clean UI
• Error Handling: Client-side notifications for upload or server failures

Setup Instructions

1. Clone the Repository

git clone https://github.com/ArshanKaudinya/PDFhelper
cd PDFhelper

2. Environment Variables

Create a .env file inside the backend folder with the following:

SUPABASE_URL=your_supabase_url
SUPABASE_ANON_KEY=your_supabase_anon_key
OPENAI_API_KEY=your_together_ai_token

Create a .env.local file inside the frontend folder with:

NEXT_PUBLIC_API_BASE=http://localhost:8000

(Replace localhost:8000 if backend hosted elsewhere.)

3. Install Dependencies

Backend

cd backend
python -m venv .venv
source .venv/bin/activate  # or .venv\Scripts\activate on Windows
pip install -r requirements.txt

4. Run Locally

Backend

cd backend
uvicorn app.main:app --reload

Frontend

cd frontend
pnpm dev
# or
npm run dev

Deployment

• Frontend deployed on Vercel
• Backend hosted on Railway
• Storage handled by Supabase buckets

Key Features

• Upload PDFs directly from your browser
• Embeddings are generated and stored locally
• Ask natural questions about uploaded PDFs
• Search within chat by keywords
• Fully responsive design with mobile support

Functional Overview

1. Upload a PDF document.
2. Ask any question related to the uploaded document.
3. Receive concise and accurate answers powered by the Mixtral LLM.
4. Search across previous conversation messages.