Lore
Lore is a document Q&A tool built from scratch as a proof of concept for Retrieval-Augmented Generation. You upload a PDF or image, and then ask questions about it. Answers are streamed in real time and cite the exact source passages.
Exploring the app
My role
Solo project. Designed and built the full stack: chunking strategy, embedding pipeline, vector search with pgvector, streaming API, and the chat UI. Also handled deployment on Fly.io including the standalone Next.js build and persistent volume setup.
Context
Built to understand RAG deeply before applying it in larger projects. The main challenges were chunking strategy (semantic vs. character-based), streaming responses without breaking the UX, and solving the Next.js standalone mode limitations with uploaded files on Fly.io.