I’m William Jeffries. I build production agents, evaluation systems, and infrastructure across healthcare, education, research, and public-interest technology.
Selected work
My best work sits where model behavior, product requirements, and operational reality collide.
Healthcare AI
A production agent that reads patient documentation and determines whether durable-medical-equipment requests meet payer requirements.
Led the core AI service from architecture through deployment, including model selection, infrastructure, stakeholder work, and offline evaluation against golden datasets.
Detailed walkthrough available on requestResearch infrastructure
A multilingual research system over the digitized holdings of the Józef Piłsudski Institute of America.
Built the ingestion, transcription, search, agent, interface, deployment, and observability systems. The agent can inspect original scans and keeps absence claims tied to document-level search evidence.
Detailed walkthrough available on requestAI production systems
AI workflows for producing quizzes, roleplays, cheat sheets, surveys, and other learning materials at course-catalog scale.
Built production pipelines, evaluator loops, and restartable browser automation that connects local operations with Google Drive, Trello, and Udemy.
Detailed walkthrough available on requestPublic-interest technology
A civic technology nonprofit that used web-connected temperature sensors to help New York City tenants document unsafe housing conditions.
Co-founded the organization and served as CTO, building its original software and sensor infrastructure around the working practices of tenants, organizers, and legal-aid lawyers.
View public repositoryHow I work
I’m a product-minded engineer who follows a system from the first user conversation through models, code, infrastructure, evaluation, and operations.
I began my career at PwC, taught myself to code, and spent the next decade moving between startups, embedded consulting teams, and public-interest technology. That path made me comfortable learning unfamiliar domains quickly and accountable for what happens after a prototype meets production.
2025–now
2020–2025
2014–2020
2010–2014
Public record
A few things you can inspect without taking my word for them.
Open source
Conference talk
Podcast
Writing
Contact