Showcase: an open-source RAG eval harness

Built a small eval harness for retrieval-augmented generation that scores answer faithfulness and citation accuracy across providers. Runs on Supabase + pgvector. Looking for contributors and feedback on the scoring metrics!