zardar khan
i build things because i'm curious and think why not.
right now that's ml systems in healthcare—messy clinical data, tight timelines, real consequences if things break.
research software engineer at sunnybrook in toronto. the nlp person. if it involves clinical text, it ends up on my desk.
i learn by building. i see a paper, i implement it. if it breaks, i figure out why. i reverse-engineer undocumented systems because that's usually where the real problems are.
notable work:
mammography classification — 50 labels → 90% accuracy, active learning loop
constrained decoding — 180x speedup, 100% schema validity on 17k pathology reports
caring contacts rag — concept to clinical trial in 3 weeks
phi de-identification — 45% → 90% F1 via synthetic data
fhir extraction — 120 days → 4 hours, found 500k missing reports
all projects →side projects:
pronunciation app — llm-as-judge for my daughter's reading
roblox game guide — taught her game dev, documented the process
sec form d job board — turned sec filings into a job hunting pipeline.
raro implementation — tried to replicate it. it got worse. learned why.
bookkeeping agent — built tools from scratch. found my $70/month starbucks habit.
all side projects →looking for remote ml roles in healthtech or applied ai.