Upload a PDF, ask questions, get answers with page citations. Streaming.
I built every piece of the RAG pipeline from scratch — PDF parsing, recursive chunking, embedding via OpenAI, FAISS vector search, and Claude for generation. No LangChain. There's also an embeddable chat widget using Shadow DOM that you can drop into any site with one script tag.
Claude, OpenAI Embeddings, FAISS, FastAPI, AWS S3