Focused notes on AI system design
Short technical essays on retrieval, domain intelligence, evaluation, reliability, and the engineering choices that determine whether AI systems work in production.
Research Note
Why Model Performance Does Not Determine System Reliability
Benchmarks are useful, but production AI reliability depends on retrieval quality, validation, observability, fallback behavior, and evaluation across the full system.
Research NoteFrom RAG to Domain Intelligence
Why retrieval-augmented generation is only the first layer of production AI, and how extraction, validation, provenance, and domain reasoning turn retrieved text into usable intelligence.