“The information gap between learner disengagement and institutional response costs 1 in 5 students their degree. SomaAI closes it.”
We build on peer-reviewed learning science and validated datasets. Every model claim is backed by reproducible evidence — not product instinct.
We treat FNR gap limits as hard engineering requirements. Demographic parity is validated before any model goes near a real student.
Every feature starts with the question: does this actually help the learner? The institution is the customer; the student is who we ultimately serve.
Our fairness audit reports are public. We publish model performance, demographic gaps, and known limitations quarterly — accountability requires transparency.
Former ML researcher at Ghana-MIT who watched too many brilliant students fall through the cracks of overcrowded institutions.
Built distributed systems at Andela and Flutterwave before turning his attention to predictive learning infrastructure.
Taught computer science at Ashesi University for six years — she understands the instructor’s perspective from both sides of the desk.
First dataset acquired: OULAD (Open University Learning Analytics Dataset). Initial GAT prototype trained on 32,000 student journeys. Dropout prediction AUC: 0.64.
64 learners. 3 courses. 1 instructor. EdNet and MoocCubeX datasets added. Multi-source training begins. First at-risk intervention triggered manually from model output.
3 institutions onboarded. 1,200 active learners. First automated at-risk alert sent to an instructor. SomaAI prevents its first documented dropout — 6 weeks after the first signal.
12 institutions. 4,821 active learners. First public fairness audit report published. GAT model v3.2 achieves 78% test AUC. FNR gap holds below 5pp across all demographic strata.
30-minute demo. No commitment. Deployment feasibility assessment included.