Data Scientist & AI Engineer, Washington, DC
I build data and AI products. Three are live:
- Justice Lens: an independent publication testing whether DC policing is applied equitably. Ten statistical analyses on 1.4M+ public police records. FastAPI, PostGIS, 420+ automated tests.
- Lease Screener: reads your lease, flags clauses that may break DC or Virginia tenant law, and answers questions about it. Two Claude Opus generator passes and a Sonnet judge; F1 72% vs. 31% for Claude Opus alone on a 4-lease test set, with 9 false alarms vs. 51. The pipeline is private; the eval is open.
- Athro News: a news digest that condenses 20 feeds into bullet points with Claude.
The code for these is private. Happy to walk through any of it.
One method, applied city by city and then nationally, with a follow-up on what the numbers measure. Read in this order:
- 59 large US cities, 2022 to 2025: Black women's police-recorded assault rate against Hispanic, White and Asian women's, in every city whose police report complete data (code)
- What that number measures: the national victimization survey, hospital records, and 911 calls against recorded victims in four cities (code)
- Nine cities side by side, 2020 to 2025 (code)
- Single cities, each with its own checks: Los Angeles (code), Washington, DC (code), Baltimore (code), Dallas (code)
- disparity-kit: the Claude Code skills every analysis above was built with, from a raw incident file to a published, caveated page
- Soccer Prediction Model: predicting non-penalty goals, 8 models compared (code)
- DC Felony Arrests 2016: interactive arrests dashboard (code)