Start here

DataDinosaur is for architects, platform owners, and data engineers putting AI into production, especially in regulated industries. Most posts write up something I built and ran: what was built, what was measured, and what the test does not show.

Read a series

The five most-read posts

  1. An Alternative to Buying Capacity to Fix Dataflow Stockouts

    A Dataflow job that can't get its machine type sends most teams straight to Compute Engine reservations. There's a free config change that went GA in April and might work better.

  2. Thinking of an agent as a Colleague

    I was challenged to work out what it would take to think of agents as colleagues. Taken seriously it becomes an operating model: identity, scoped access, governed onboarding context, supervision, evaluation, deprovisioning. Here's what worked when I built that, and the three places it broke.

  3. Multi-turn context safety alerting: catching what a per-message filter cannot

    Per-message filters like Model Armor correctly pass most of a multi-turn probe. Adding the conversation tier that catches it, and the alerting fallback that turned out to matter more.

  4. Managing Data Products in Google Cloud's Knowledge Catalog

    Dataplex's Knowledge Catalog turns data products into first-class, discoverable, requestable objects. A full field tour — finding, describing, sharing, and securing certified data products — plus the gotchas the docs skip.

  5. Stop Your GCP Bill Before It Stops You

    AI has made it almost too easy to spin up GCP services without thinking through the cost implications. Here's how I built a hard cap that actually cuts off spending - using Google's own recommended pattern.

Other ways in