Software engineer and founder of Aevox, AI analytics for metabolic testing deployed in five labs. I build AI software end to end and work forward-deployed: the first sales call, each lab's raw exports as they actually arrive, and the fix that ships the same week. Exercise physiologist (MSc Applied Physiology, UCCS), NYU Computer Science, three years building analytics and trading credit at Barclays.
Aevox's code is private. How it's built: How Aevox reads a metabolic test covers ingest under physics checks, a deterministic threshold engine, a bounded model read, and a physiologist who confirms.
- motrpac-redox: multi-omics analysis (R and Python) of the MoTrPAC endurance-training dataset, behind a manuscript in preparation
- wearable-validity-atlas: computed validity grades for consumer-wearable claims, from the independent literature (live)
- longevity-biomarker-heatmap: 135 biomarkers, 519 citation-backed evidence cells (live)
- bioML: metabolic prediction from wearable signals on a synthetic cohort
- podex: a local-first daily audio briefing from an X feed (Claude + ElevenLabs)
Also: Parsel, a voice digital biomarker for Parkinson's, winning entry at the Healthcare × AI NYC Hackathon 2026.
Resume · jackmislinski.com · LinkedIn · X