I build cool things.
Computer science and public health student at Stetson University.
Selected work · 4
Understand customer complaints—and see what informed the prediction.
Find opportunities. Keep track of what happens next.
Make years of clinic records easier to understand.
See the risk behind the return.
Selected work
01 · Machine learning · Evaluation
Understand customer complaints—and see what informed the prediction.
02 · Systems · Automation
Find opportunities. Keep track of what happens next.
03 · Data · Public health
Make years of clinic records easier to understand.
04 · Backend · Financial data
See the risk behind the return.
Selected work
I’m the only contributor to all four repositories. Each has a live demo, its code, and a write-up of the decisions behind it.
01 · Machine learning · Evaluation
Understand customer complaints—and see what informed the prediction.
Classifies complaint themes, highlights contributing words, and routes uncertain predictions for review.
Review where it helps. I chose the 0.40 review threshold on 2019 validation data. It flags about 36% of the 2019 test set and catches about 63% of its errors. An earlier 0.65 rule flagged 76% of records, which made it useless for triage.
02 · Systems · Automation
Find opportunities. Keep track of what happens next.
Screens internship postings and brings application stages and email updates into one workflow.
Explainable scoring over ML. Scores come from a deterministic rule engine, so every score and every rejection carries a stated reason I can audit.
03 · Data · Public health
Make years of clinic records easier to understand.
Reconciles inconsistent exports into readable reporting. The public demo uses synthetic data.
One canonical model. The pipeline handles mixed date formats, diagnosis codes vs. free text, and “Gender Identity” vs. “Gender” headers explicitly, rather than assuming every year has the same columns.
04 · Backend · Financial data
See the risk behind the return.
Explores historical performance, drawdowns, volatility, and correlations for a selected stock portfolio.
Refuse to fill gaps. An interior gap returns MISSING_DATA instead of being dropped or forward-filled. Dropping the date would turn a multi-day move into one “daily” return.
About
In each role below I started by learning what the people using the data needed. Then I built the smallest thing that answered it, tested it, and wrote down where it stops being reliable.
AI is part of how I build. I use coding agents for implementation and code review, and I run them inside harnesses with feedback loops so they check their own work and each other’s. For a bug fix, the agent first writes a test that reproduces the bug and fails, so it can’t claim to fix something it never reproduced.
Nothing is accepted on trust. I review and test every change myself, rely on end-to-end tests that show real behavior, and remove the low-value tests agents tend to pile up.
Worked with clinic staff to identify reporting bottlenecks and define requirements for privacy-conscious reporting. I also supported data analysis, documentation and patient intake. The campus health dashboard came out of this work, and staff used it to review data that had only existed as raw exports.
Migrated more than 250 paper-based records into an SQLite database, improving access speed by 25%. I built a Python system with automated validation checks and audited the migrated records against the source documents, reaching 99% field accuracy.
Lead coding and technical learning sessions for Black and Latinx computer science students, and share the tools and practices I find useful.
Minor in Civic Engagement. Bonner Scholar (full scholarship and service award), Dean’s List.
Contact
