Products is the line that moves the unit's research into deployable tools — primarily, in the first phase, for African financial systems. Open methods, paid integrations, real data.
The first cohort of products sits in credit scoring, fraud and anomaly detection, and customer-analytics tooling for banks, microfinance institutions, and mobile-money operators. The unifying technical bet is that interpretable, mathematically-grounded methods can match or beat black-box baselines on African banking and mobile-money data, once you take feature engineering seriously and stop pretending the data looks Western.
Two patterns we keep finding: time series in mobile-money settlements have non-trivial topological structure that classical features wash out; and credit data for semi-formal income earners carries information in its missingness that needs interpretable handling, not blind imputation. Our products are built around what these patterns enable.
Every product traces back to a published or in-publication paper from the Research line. If we cannot point to the paper, we are not selling the method as ours.
Tools, libraries, evaluation benchmarks, and anonymized worked examples live on GitHub. Trained models built on client data stay closed; the methods that produced them do not.
We test on actual African banking, mobile-money, and microfinance data, with its missingness, its intermittent labels, and the regulatory constraints (BCEAO, BCBS, EU AI Act for European-facing institutions) that make some standard tricks unusable. Benchmarks that score well on US credit datasets do not earn a pass here.
The split exists for a reason. Reputation compounds only if methods are public. Revenue exists because banks pay for outcomes, not for ideas. Both halves are needed; collapsing either kills the institution we are building.
A 1-page brief of the problem and the data you can share: contact@airina.africa
See repositories at github.com/AIRINA-Labs
Scaling Phase-1 work into Phase-2 (agriculture) earlier: contact@airina.africa