ENGINEER / ARCHITECT OF SYSTEMS / APPLIED AI
AVAILABLE WORLDWIDE
I MAKE SYSTEMS BEHAVE.
<Try to Inspect Me>
The API looks fast. The workflow looks simple. The agent just does the thing. Underneath, there is usually a mess of decisions, bottlenecks, retries, permissions, and edge cases making that possible.
That is the part I like. Find what is actually getting in the way. Fix it properly. Then make sure the next person does not have to rediscover the same answer.
Most of this started with something that was too slow, too fragile, hard to validate, or doing far more work than it needed to. I went underneath, found the real constraint, and changed it. These are the parts you can measure.
Led inventory-processing optimization across approximately 2,000 stores. Full-feed completion fell from ~21 days to ~2 days, while database CPU fell from ~90% to 10–15%, without the previous slowdowns in dependent applications.
In a separate production investigation, reduced timeouts from ~20% to near zero in an inventory flow receiving 10K+ Pub/Sub messages per second at peak, using bounded concurrency, flow control, capacity tuning, and targeted retries.
Built SIM-E2E, an integration-testing platform adopted across multiple teams. Story-linked suites bring together automated API, BigQuery, and Spanner checks, so teams can validate more in each run.
Built a platform that scores multi-turn conversations in parallel on Apache Beam and Dataflow, with a self-service console for datasets, evaluation runs, saved configurations, and run history. Teams can create golden datasets, expand them with generated scenarios, and continuously re-run evaluation suites as models and prompts evolve. At Scale.
01 / Curiosity
When something is slow or keeps breaking, I want to understand what is actually happening. I tend to keep digging until I can explain it.
Scroll to explore