STAFF ML ENGINEER / AI SAFETY
Engineering
for AI safety.
I build evaluations that expose failures—and help engineering teams use them to decide what ships.
I’m looking for engineering work in AI safety.
01 / SELECTED WORK
Build it. Test it.
Question the result.
Production systems, independent experiments, and the work of getting people to use the evidence.
ENGINEERING LEADERSHIP · BOOSTLINGO
Make evaluation part
of how a team ships.
I drove the shift to eval-driven development: shared tooling, teaching, and reliable checks that became part of the release process.
BEHAVIORAL EVALUATION · ADVERSARIAL TESTING
What does the test actually measure?
Spoken role-redirection tests, hidden instructions in résumés, and behavioral audits with Inspect and Petri.
INDEPENDENT ML EXPERIMENTS · AI-ASSISTED
Catch harmful requests.
Count the false alarms.
Comparing TF-IDF and DeBERTa, investigating errors, and checking what survives reproduction and evaluation on other data.
PUBLIC RESEARCH · MEDICAL INTERPRETING
Measure meaning when it matters.
Designed and authored a medical AI interpreting accuracy study. Co-presented a tutorial on spoken-AI evaluation at AMTA 2026.
02 / THE THREAD
Big questions.
Practical work.
I started with political science and philosophy. Discovering AI safety in 2016 gave those interests a direction. I pursued ML, built products, and learned how much depends on asking the right questions of a system.
How I got here03 / EARLIER WORK & OTHER CURIOSITIES
A few more threads.
Love letters & learning machines
Siamese SBERT, historical authorship, and the limits of a convincing score.
BIAS / EXPLORATORY ANALYSISChess after The Queen’s Gambit
What sentiment analysis can miss about discussions of women in chess.
EARLIER NLP RESEARCHTrue Love or Lost Cause?
Stylometry and neural models before the SBERT study.
2017 / GENERATIVE MODELINGOne character at a time
An early Keras LSTM text-generation experiment. Uploaded for posterity in 2020.
And some things I built along the way
04 / WHAT’S NEXT
Let’s make AI
worth trusting.
I’m looking for hands-on engineering work in AI safety: behavioral evaluations, adversarial testing, and research infrastructure. Open to discussing scope and level with the right team.
zacbolton2129@gmail.com