Blog

Writing on AI & automation

Notes on AI engineering, automation platforms, and building production systems.

  • There's no single "best" RAG approach

    Naive RAG, hybrid search, re-ranking, and agentic RAG each solve a different failure mode — picking the right one starts with knowing what's actually breaking.

  • Hallucination isn't a bug you fix once

    Why hallucination is a property of how LLMs generate text, not a defect — and the layered mitigations that actually reduce it in production.

  • Handling multiple files in a RAG pipeline

    Why naive multi-document RAG breaks, and the practical fixes — metadata filtering, per-document chunking, and source attribution.

  • Welcome to my blog

    Why I'm starting to write in public about AI engineering and automation.