15+ years building data-intensive applications. Currently focused on agentic systems, AI observability and evaluations.
The visual search I built in 2018 had about 1.5 million uses a week. Even partitions gave hot clusters, because per-item compute was expensive.
About a third of production models could not be recreated when I took over data science and data engineering. We put the ML pipelines under CI/CD.
quickmbti.com, which I built on my own, has about 195,000 test completions since Aug 2025.