January 2025data scientist
The measurement that mattered
Nexus is a decision intelligence platform for K-12 school districts. That means a district's data has to arrive correctly and on time before anything else we build has any value at all. When I joined as a data scientist, the nightly integrations that feed every one of our products were outgrowing the architecture they had originally been built on, and keeping them running was consuming a large share of our most senior engineering time. That turned out to be the measurement that mattered. The rate at which we could bring on a new district was also the rate at which the company could grow.
Summer 2025data scientist
Rebuilding the part that set the rate
Over the summer I designed the ingestion platform that every district now runs on. The constraint I set was that somebody other than me had to be able to onboard a district with it. That meant the vendor-specific knowledge had to live in configuration rather than in code, and every change had to prove itself against production data before shipping. It became the standard path for the whole integration team, which is what actually moved the rate. The senior engineering time it gave back went into building product instead, and in October I became Head of Data.
Early 2026head of data
Correctness as a property of the system
A data company really only sells one thing, which is the assurance that the number is right. In order to make that true by construction rather than by vigilance, I built the layer that watches the pipeline itself. It notices when a source has quietly stopped updating, it distinguishes a legitimately empty file from a missing one, and when a job does fail it identifies the cause, drafts the fix, tests that fix against the data that broke it, and hands a person a change to approve. The diagnosis is done by the system and the decision is kept by a person, which is the arrangement I would like everywhere. In January I became Senior Systems Architect.
Summer 2026systems architect
Putting the data to work
Once the data could be trusted, it could start doing work rather than simply being reported on. District leaders now read board and principal dashboards every morning. A transportation bidding process runs end to end, because the coordinator who had been doing it by hand asked for a system that decides rather than one that drafts. A graduation-likelihood model is calibrated carefully enough that a district can defend it in front of its own school board. Underneath all of that, I spent a good deal of the year on the platform itself. I split the nightly run across two pods and two queues on Kubernetes, added an admission gate that bounds total memory rather than raising the limit, and rewrote the most expensive paths, including one transform that peaked at 16 GB and now peaks at 2.4 GB. Treating Azure cost and capacity as engineering targets, rather than as a budget line somebody else owns, is what let the fleet keep growing on the hardware we already had.
Nowsenior systems architect
The instruments pointed at Nexus
Most recently I built the platform that Nexus uses to see itself. It pulls every piece of client work out of the four systems we track it in, sweeps the team's calendars each night to attribute client meetings by attendee domain, and links work items to each other and to the meetings they came from. On top of that sits a weekly review which puts person-days per client against what that client pays, with the methodology printed on the page so that anybody can argue with it. Building the reporting turned up two problems in the underlying data, including a thirty-second network blip that had quietly dropped about twenty person-days. A company that can see itself makes noticeably better decisions, and I have really enjoyed watching ours start to.
