26 May 2026
Offline eval: why I test ingestion before I test reasoning
How AMICOVERED's regression battery works across three layers — document ingestion, retrieval, and inference — and why the order matters.
Writing
Longer notes on products and engineering decisions.
-
-
6 May 2026
Debugging journey: from naive RAG to an agentic loop
How AMICOVERED’s architecture moved from a standard RAG stack to a staged pipeline, then to structure-aware chunking, and finally to an agentic retrieval loop—driven by two different kinds of failure.