At FlyRank, I built Python command-line pipelines that enforce data contracts, block target leakage, and create an audit receipt for each run. In my capstone, I analyzed more than 100,000 pages from about 50 clients and found that a random split overstated model accuracy compared with holding clients out.
I also audited data quality, finding 6,390 duplicated records in a sample table, and documented their effect. I wrote a ranked action playbook with reason codes, human-review steps, and monitoring triggers.
For my Content Decline Risk Analysis project, I validated a random forest using client-grouped splits. A two-code review rule separated pages that declined 68.0% of the time from pages with neither code at 60.2%; I published the paper with limitations and ranked recommendations, without client-identifying data.
My other projects include an Organic Content Refresh Triager that narrowed 30,000 URLs to 2,070 high-priority candidates, and a diabetes risk model comparison where class-weighted LightGBM achieved a ROC-AUC of 0.827. I’m an MPH candidate in epidemiology, and I support project workflows and maintain structured tracking sheets at the Anambra State Signage and Advertising Agency (ANSAA).
