How vedbabar/chat-with-pdf works
Project type and purpose
• This is a full-stack, multi-tenant web application (“Chat with PDF AI”) that lets authenticated users upload PDF documents, then chat with an AI agent (Google Gemini) about their content.
• It comprises a React/Next.js frontend, a Node/Express backend, a background worker, and three data stores (PostgreSQL via Prisma, Redis for queuing, Qdrant for vector search).
File structure
• Top-level:
– client/ (Next.js + TypeScript + Tailwind + Clerk for auth)
– server/ (Express API in index.js, Prisma schema & migrations, file upload directory, a worker.js)
– docker-compose.yml (orchestrates PostgreSQL, Redis, Qdrant)
• Patterns:
– client/app/components → UI and page components (chat view, file upload, sign-in/up)
– server/prisma → database schema and migrations (users, chats, file metadata, vector references)
– server/uploads → persistent PDF storage
– server/index.js → HTTP API endpoints (authentication middleware, file upload, chat management, question endpoints)
– server/worker.js → BullMQ + Redis job consumer, handles PDF parsing, embedding and vector insertion via Qdrant
README for additional insights
• Auth: Clerk (publishable + secret keys)
• Background Processing: BullMQ + Redis
• Vector Search: Qdrant
• AI: Google Gemini via LangChain.js
• Database: PostgreSQL via Prisma
Detail
• Environment variables or config points next to each component (e.g., .env for API, .env.local for Frontend).
• Port numbers (3000 for Next.js, 8000 for Express).
• Docker services for Redis, PostgreSQL, Qdrant from docker-compose.yml.