Projects
Four things built from scratch, each with the number that mattered.
13 sto reproduce the whole study
The implementation behind the paper. A typed lineage DAG, a blast-radius query, and a recovery planner that classifies every affected model by whether a clean, signed ancestor is reachable. About 1,500 lines, and a checker validates every number in the paper against the published results.
30,000+live metrics
StackStorm is an open-source automation platform used across cloud data centers. I built integrations piping live operational metrics from Telegraf monitoring agents into the platform in real time, and contributed the improvements upstream.
End to endon AWS
A classifier that distinguishes human writing from generated text, trained and deployed on AWS from raw data ingestion through to live prediction. Built in response to concerns about generated content in academic settings.
100,000+events a day
An API for large-scale event ingestion and analysis, designed to scale horizontally with rate limiting and background job processing so nothing blocks the request path.