Writing
Technical deep dives on LLM systems, ML engineering, RAG pipelines, and building production AI. Written from experience, not theory.
I have built several RAG systems. The first looked great in demos. It fell apart within a week of real usage. The failure modes are predictable - and fixable. Here is what goes wrong.
At Omdena, I moved a classification model from 72% to 89% accuracy without changing the model architecture. Here is the methodology: ablation first, architecture second.
I built LUNA because I was tired of API costs and logging. Six months later it routes across 8 LLM providers with a plugin system, persistent memory, and on-device voice. Here is what I learned.
Monthly. No spam. Unsubscribe any time.