A release gate for three product lines and one team
Designing a test framework that gives release managers a single pass/fail signal.
Coming soonInsights
Notes from the work: what held up in production, what didn't, and the numbers behind both.
Why most RAG systems fail at retrieval, not generation
What we learned tuning a 55-million-object vector cluster, and the metrics that predicted answer quality.
Read articleHow much autonomy should an agent have?
A practical framework for deciding what agents may do alone, what needs approval, and how to expand trust.
Read articleGradient boosting still wins on tabular data
Notes from building credit-risk ensembles that decide in under 100 milliseconds.
Read articleComing next
Offline-first mobile: what actually breaks in the field
Sync conflicts, background limits and the patterns that survived a last-mile delivery rollout.
Coming soonModernising BI without moving the data
How a dbt semantic layer over an existing warehouse ended the spreadsheet wars.
Coming soon