Hey Product Hunt!

I built ReportIQ while exploring a question that goes beyond building another AI chatbot:

How do we know whether a RAG system is retrieving the right evidence, rather than just generating a convincing answer?

ReportIQ is my attempt to explore that problem through a working document Q&A application, retrieval evaluation, and LLM observability.

What you can explore:

Ask questions about PDF documents.
Explore a RAG pipeline built with BGE embeddings and FAISS.
Evaluate retrieval quality using metrics such as Recall@K and MRR.
Inspect LLM workflows using LangSmith.

I processed 8 PDFs into 721 chunks and evaluated retrieval on 50 questions. My latest recorded baseline achieved 91.7% Recall@3 an initial result from a evaluation set, not a claim of general performance.

Built with Python, Streamlit, FAISS, BGE embeddings, and LangSmith.

Try ReportIQ: https://reportiq.streamlit.app/

Explore the code: https://github.com/Ronit-019/ReportIQ

I'd especially appreciate feedback on the retrieval evaluation approach, the user experience, and what you'd improve next.

Thanks for checking it out!