Founding Engineer · Data Infrastructure × AI

Building platforms where agents can be trusted with enterprise data.

I lead architecture and engineering for an agentic data fabric platform: ingestion, storage, transformation, semantics, and serving on the data side; retrieval, agentic systems, evaluation, and ML inference on the AI side. The interesting problems live where those two halves meet.

Selected work Writing

Writing

One capability, four front doors Jul 2026 The number carries its own receipt Jul 2026 The only implementation of our budget was named NoBudget Jul 2026 All writing

Work

00 Trusting agents with enterprise dataAn execution model for autonomous agents that survives enterprise security review. 00 How Removing Docker Taught Me That My Tests Were Lying (And So Was I)Removing Docker from your test harness changes OS semantics you never declared, and two passing tests went green because of it. 00 Your gold set is lying to youReference answers are artifacts with bugs, and they are the only software in the stack that nobody code-reviews. 00 Retrieval you can auditDocument intelligence where every answer carries its provenance. 00 Fifteen services, one installDelivering a fifteen-service data platform into customer Kubernetes environments in one step. 00 When the spec is too big to generate fromA subsystem's OpenAPI spec was so large and self-contradictory that generating a client from it became a runtime liability, so I hand-built the types instead. 00 The memory no one reviewsThe agent forgets everything between runs, so I let the harness mine each run and write lessons into the next one's instructions, and the hole I did not see is that nothing reviews a rule the machine writes for itself. 00 Four reviewers, one mindI let an AI agent merge to main when a panel of AI reviewers unanimously approves, and the slow realization was that unanimity from one model is not four opinions, it is one opinion read four times. 00 What regulation does to architectureYears of healthcare engineering taught me that compliance, taken seriously, is a design input that produces better systems. 00 Managing engineers who never sleepA year of running AI coding agents turned into an accidental management apparatus, written one rule at a time. 00 Inducing the schema instead of supplying itThe hard part of turning a pile of documents into a database is not filling the tables, it is deciding what the tables should be, and the rule that made mine work is that a column earns its place only by paying for itself across the whole corpus. 00 Robots that made burgersDirecting the software for autonomous burger assembly at Momentum Machines. 00 A lockless hash table in a database coreInside SAP HANA, where a single data structure tripled parallel query throughput.

About

Fifteen years across database cores, robotics, big-data platforms, healthcare engineering leadership, and two startups: one acquired, one mine. Now a founding engineer betting that the next platform layer is the one that lets AI act on enterprise data safely.

The longer arc