innerdance — a RAG system you can trust to cite its sources
Retrieval-augmented Q&A over a large, mixed corpus — with measured answer quality.
The problem
Answering questions over a large, mixed corpus (audio transcripts, PDFs, EPUBs) where a confidently wrong or unsourced answer erodes trust immediately.
The approach
The complete pipeline: multi-format ingestion, token-based chunking, embeddings in PostgreSQL/pgvector, hybrid retrieval (semantic + keyword search fused with Reciprocal Rank Fusion, plus reranking), exposed as tools to a Pydantic-AI agent that streams answers grounded in cited source passages.
What it proves
An evaluation framework with LLM-as-judge scoring, a custom error taxonomy, and Langfuse observability — answer quality is measured and improved iteratively, not guessed. Every answer is grounded in cited source passages.
Highlights
- Cited sources on every answer
- LLM-as-judge eval harness + error taxonomy
- Hybrid retrieval: RRF + reranking
- Langfuse observability
Have a process or product that should exist as software?
If you bring the domain expertise — or you're a seed / Series A team that needs strong hands — I'll handle the technical side end to end, or work with your team to ship a product people love to use.