Isolate the thing that varies.
I build systems where the part guaranteed to change — a TTS model, a market venue, a clock — sits behind one interface, so everything above it cannot tell the difference. Currently tech lead for platform observability at Capital One (5,000+ applications) and founder of PrimeMaxing. Each page here answers four questions, in order: what was the problem, why was it hard, what I decided, and what happened.
The work
- PrimeMaxing 2024– · founder · in pilot AI-driven blood biomarker analysis for clinics and patients — architecture built so patient data can never leak into the parts that don't need it.
- Crypto Arbitrage Monitor · in progress live monitor · paper only · coming A live two-venue BTC top-of-book monitor with fee and size controls — built to show, interactively, why retail crypto arbitrage doesn't work.
- Algo OMS · in progress python · production path · replay coming An order-management system whose backtest runs the production event path — and, soon, a nightly replay you can watch here.
- Echo personal infra · in daily use Batch-generate narration locally on a GPU, timestamp it for follow-along highlighting, and stream it anywhere through a small password-locked reader.
- Observability at Capital One 2025– · tech lead · 3 teams, 18 engineers OTel pipeline at 100–300 TB/day, a monitor migration that deleted 99.97% of the monitors, and an AI incident-analysis platform that halved resolution time.
- Contract work Centennial Labs · Woodford House Lab-result triage under the COVID surge, and an 18.7 GB corpus analysis for a school productising its mentoring programme.