Now
This month
Chicago, working globally. Updated September 2026. Short, on purpose.
Building
- An AI-augmented ebook QA pipeline that has to survive a storefront release, not a demo day. Models propose; humans still sign.
- Nexacore as a venture vehicle for an AI quality engineer that other teams can actually run — oracle, audit trail, escalation rules.
Thinking about
- What a quality contract for an agent looks like in writing. Not a prompt. A policy: what it may change, what it must escalate, what evidence it leaves.
- Whether “eval engineer” should sit in QA or in the model org. I think QA. The people who want the eval to pass should not own the eval.
Reading
- Empirical evaluation papers on model behavior under distribution shift — the QA problem, under a different name.
- Incident write-ups from teams that shipped LLM features without an oracle. The pattern is familiar.
Also
- House of Elango, slowly. Taste is a quality problem too: what you allow to ship under your name.