• Built a buyer-intent model (XGBoost, logistic regression) on ~37 behavioral features across a ~316M-row Presto/Hive dataset to predict first-time buyers of affiliate content products, reaching ROC-AUC ~0.79.
• Found and fixed a partition-join defect causing a 30% null rate across ~96M users, root-causing it to a pipeline bug rather than a real signal, and shipped the corrected pipeline to production.
• Validated model retention: high-intent users retained at 1.7× low-intent and captured +1.75M incremental buyers at fixed spend, projecting $117–175M in annual GMV and setting the targeting cutoff for an A/B test designed with power analysis.
• Surfaced 5 commerce signals as the top intent differentiators and presented the findings to the discovery and ranking team, who added these signals to organic affiliate ranking.
• Measured expected CTR impact of six consumer UX levers from observational data, moving three high-impact levers into execution, deprioritizing two, and getting a proposed social-proof lever added to the roadmap.
• Delivered 8 product-category priors on GMV-per-1,000-views efficiency for high-intent users, now used in offline ranking, and sized 5 pre-click metadata signals subsequently implemented by the UX team.

• Python
• SQL
• XGBoost
• SHAP
• A/B Testing

• Engineered a memory-safe, fallback-enabled streaming pipeline to run the FastReseg RNA resegmentation algorithm across all 361 FOVs of the Colon TMA CosMx dataset (447,722 cells, ~247M transcripts), preventing OOM failures at full scale.
• Built the full annotation pipeline from raw CosMx output to labeled clusters — QC-filtered to 424,423 cells, 3,000 HVGs, PCA/Leiden into 15 clusters, top-20 markers per cluster — enabling downstream DE and tumor-scoring analyses.
• Conducted pseudobulk differential expression with DESeq2 (ashr shrinkage) by cell type across healthy vs. DSS colitis conditions, generating condition-specific gene signatures that characterize colitis-associated expression shifts.
• Built a clustering workflow integrating 424K CosMx spatial cells with 826K scRNA-seq cells via Seurat SCT and RPCA, reclustering against an independent reference to refine tumor microenvironment annotation beyond vendor labels.

• Python
• R
• Seurat
• DESeq2
• Spatial Omics

Building an end-to-end ML portfolio project under industry mentorship, with weekly cohort sessions on applied ML, responsible AI, and code review.

• Machine Learning
• Responsible AI

Led recitations in Java-based algorithms and systems, designed and graded 12+ programming assignments, and tutor students one-on-one in discrete math, graph theory, and algorithm design.

• Algorithms
• Java
• Teaching

Founded a STEM outreach foundation in Yaoundé, Cameroon that has engaged 200+ high-school students through workshops in Python, SQL, Arduino, and C/C++.

• Python
• Mentorship
• Community