Hi👋🏿, I'm Blessed
Software engineer and researcher. I study how to keep AI-powered data systems correct as data, definitions, access rules and tools change, and what that costs.
Current: a paper on what agent traces do not record, a study of whether AI agents know when they cannot see all the data, and Sedge, a data operating system.
, or email me: I am open to AI and research engineering roles (remote & relocation).
Papers in journals, conferences and on arXiv.
Performance Anomaly Detection in Athletics: A Benchmarking System with Visual Analytics [ArXiv]
Published*Blessed Madukoma, Prasenjit Mitra
13th IEEE Swiss Conference on Data Science and AI (SDS2026) · Zurich, Switzerland · May 06-07 2026.
Benchmarked 8+ statistical, ML and Bayesian methods over 1.6 million athletics performances from more than 19,000 competitions, using confirmed sanctions as partial labels. The platform links each flag to supporting records so investigators can examine the evidence; a flag is a lead, not a verdict. The full paper received Honorable Mention - Best Paper Award at IEEE SDS 2026. Explore the live platform.
Unmasking COVID-19 Vulnerability in Nigeria: Mapping Risks Beyond Urban Hotspots [ArXiv]
PublishedSheila Wafula, *Blessed Madukoma
Women in Machine Learning Workshop @ NeurIPS 2025 · San Diego, California, United States · Dec 01 2025.
Built a composite risk score for COVID-19 vulnerability across Nigeria's 36 states and the Federal Capital Territory. The score combines poverty (weight 0.4), access to healthcare (0.3), population density (0.2) and age (0.1), adjusted by case rates per 100,000. Lagos scored 673.47 against a national average of 28.16 and had 35.4% of national cases. The score shows where testing, vaccines and health resources should go first. Rural underreporting means that the real risk outside the cities may be higher than the score shows.
Have a wonderful Tuesday 🚀
