Oluomachukwu Emenike

Organic content refresh triager

FlyRank AI. Python, Pandas, command line.

Traffic drops are often seasonal or low-value, so I looked for Page-1 pages with large impression volumes but far fewer clicks than their position should earn. The agent quarantines leakage fields before scoring, fills missing zero ranks with cohort medians, and writes a markdown audit receipt on every run.

It cut 30,000 URLs to 2,070 high-priority candidates. The top page sat at position 2.5 with 509,252 impressions in 90 days and a 0.15% click-through rate against a 2.71% benchmark, roughly 13,000 recoverable clicks.

Diabetes risk prediction

Python, scikit-learn, LightGBM, XGBoost.

An end-to-end pipeline on the CDC BRFSS 2015 Diabetes Health Indicators dataset (253,680 respondents). It compares five models under four class-imbalance strategies, ranks features, and audits sex and age-group subgroups for fairness. Scaling and resampling are fit on training data only, to prevent leakage.

Class-weighted LightGBM reached ROC-AUC 0.827 and caught 79% of diabetic cases on 50,736 held-out respondents. Without rebalancing, models scored a similar AUC but caught only 13 to 15% of cases.