writing
Notes from production
Lessons from data platforms that had to work — lakehouse design, batch orchestration, streaming ingestion, and putting LLMs on top of governed data.
June 20, 2026
Designing row-level security for a 1,600-user lakehouse
Eight organizations, one Iceberg lakehouse, and the governance model that let them share infrastructure without sharing data.
June 5, 2026
Idempotency patterns for Kinesis + Lambda ingestion
In financial data, a duplicate is a defect. The patterns that kept replays and retries from double-counting events.