AI4M · R² on a survey it never saw
0.29against 0.09 for the best baseline
- State level
- 0.45
- Ten states held out
- 0.23
- Labels shuffled
- Below 0, as it should be
Six years of production software, mostly in healthcare. The roles, then the evidence.
One number
Malaria risk forecasting for every state and local government area in Nigeria. I work across all of it: the system design, the data pipeline and model, the API, the dashboard, and a public docs site with SDKs.
A live platform that identifies 227 African species from photos and video, and 72 from sound. I owned the data, training, API, frontend and infrastructure, and co-designed the UI.
An electronic medical records system across 10+ hospitals and 1,000+ daily clinical users, on infrastructure I designed.
Patient queues across 15+ hospitals: multi-tenant, with an integration that ended duplicate entry at registration. I co-designed the UI and trained staff on site.
Aria, a workflow engine with LLM steps that draft replies and summarise tickets; the support platform’s ticketing, live chat and CRM; embeddable SDKs.
Location-based search connecting 2,000+ patients with healthcare providers near them.
An admin dashboard for healthcare administrators, with bulk upload for large datasets.
UI direction and information architecture for an investment platform, and its first component structure.
Applied ML
2 models in production
Each shown with how it was checked, not only what it scored.
0.29against 0.09 for the best baseline
75.9%a fine-tuned Vision Transformer
For a role, a product, or a second opinion on one. A call books straight into my calendar.