Why most AI pilots die before production
The demo worked. Everyone clapped. Six months later, nothing is deployed. The failure usually isn't the model. It's everything around it.
6 min read
Insights
Notes from the workshop
Practical writing on architecture, machine learning, and data engineering. The thinking behind how we build.
Medallion architecture, in plain language
Bronze, silver, gold: the three-layer pattern that keeps data platforms sane, explained without the vendor jargon.
5 min read
RAG systems fail quietly. Evaluate them like they will.
A retrieval-augmented system rarely crashes. It just starts answering confidently from the wrong page. Honest evaluation is the only defence.
6 min read