Blog
Notes on data & software
Software development, data engineering, and the systems behind them.
Agents · 8 Sept 2026
An agent is not a DAG with vibes
Agentic workflows look clever in a demo. In production they need the same contract as a pipeline: tools, retries, and a human who can stop them.
LLM · 19 Aug 2026
LLMs on your warehouse, not a chatbot in the way
Text-to-SQL without a grain, a catalog, and a kill switch is a hallucination with a login. Here is how we wire models to tables you already own.
AI · 22 Jul 2026
AI without the enterprise circus
Most AI programs fail for the same reason data lakes fail: no owner, no grain, no legal basis. A thin slice beats a platform slide.
Pipelines · 12 Jun 2026
Late data, retries, and the boring parts of a pipeline
A pipeline that only works on happy-path batches is a demo. Here is how we treat late arrivals, retries, and alerts as the product.
Architecture · 6 May 2026
System design for a team of five, not a platform org
Good architecture is the contracts you can operate at 2am. Here is the shape we use on GCP and AWS when nobody has a dedicated platform squad.
Modeling · 4 Mar 2026
Warehouse models your product and finance people can query
A warehouse nobody understands is a lake with extra steps. Grain, names, and one place for the metric.
Analytics · 18 Nov 2025
Boards on your database, not another SaaS silo
A dashboard you cannot leave is a tax. DaaS should sit on your warehouse and disappear if you fire us.
Healthcare · 21 Aug 2025
A patient is not a user id
Healthcare pipelines fail when you copy the SaaS identity model. Consent, retention, and purpose have to live in the data.