Written for an engineer reading the night before interviewing me: what the system did, how it was put together, and the parts I got wrong. There is also a CV.
An autonomous browser agent that runs a product's real user flows on every PR and reports failures with evidence from an actual execution. Four technical founders, no sales team, deals up to $50K ACV, and a weekly regression pipeline at 98% accuracy for a major Argentine bank.
Read the write-up →Real-time deep-learning analytics over ~400 airport CCTV cameras, air-gapped, sub-second, on two physical servers.
Read the write-up →The largest single pipeline ever run through Google's Vision API: 350M images in a two-week backfill, feeding a two-tier IAB taxonomy classifier served on Vertex AI.
Read the write-up →Systems that run where the cloud isn't: local models across a McDonald's restaurant fleet, a T5 trained inside an air gap. Plus the AI learning path for Googlers (~500 members, Initiatives Champion award) and the Universidad Austral AI lab.
Read the write-up →