Most of my career has been the part of machine learning that gets hard after the demo: real-time inference on live data, where latency, throughput, and reliability under changing conditions matter more than another point on a benchmark. Most recently I led a data science team shipping detection, classification, tracking, and segmentation into production.
For the past two years I ran an independent quantitative research program. I built the data pipelines, the options analytics, and a tick-level backtesting and validation framework, then used it to test thirteen strategies and reject twelve. The work I am proudest of there is not a strategy. It is the methodology: null controls, out-of-sample discipline, and catching the ways a backtest lies to you before you trade it.
Earlier: a postdoctoral fellowship at the U.S. Army Research Laboratory (behavior classification, motion prediction), and a PhD at the University of Georgia in spatial place recognition for mobile robots.