Ask questions to your documents. Always with sources. Or no answer.
demo-github.mp4
Larry processes PDF documents and answers questions based on their content, always providing sources or explicitly stating when no answer can be found.
Architecture:
- Web: Next.js application for document upload and chat interface
- API: Node.js REST API handling authentication, document management, and RAG-powered question answering
- Worker: Node.js service consuming SQS messages to process documents asynchronously (parsing, chunking, embedding generation)
Infrastructure:
- PostgreSQL (+ pgvector) - Data storage and vector similarity search
- Redis - Caching and real-time progress updates
- MinIO/S3 - File storage for uploaded PDFs
- LocalStack / AWS SQS - Message queue for async document processing
- OpenAI API - Text embeddings and chat completions
- Quick Start - Get Larry running locally in 5 minutes
- System Workflows - Diagrams of main flows
- API Reference - Available endpoints
- Roadmap - Next steps and future plans
- Contributing - How to contribute
- Technical Specs - See
/docsdirectory for product and technical documentation