Cohort 1 TDA for Data Scientists opens this quarter in Cotonou. Apply →

The shape of African data.

AIRINA is a research institute in Cotonou, founded in 2025. We work on topology and applied machine learning, applied to the questions West African banks and microfinance institutions actually ask. We publish before we claim. The institutions that pay our bills are the ones that send us their problems; the trainees do not pay.

Cotonou, Bénin  ·  Phase 1 of 3 — finance, agriculture, health.

What brings you here?
01 · You build credit

A bank or MFI

Looking for credit scoring, anomaly detection, or risk tools deployable in the BCEAO zone.

See the products →
02 · You work on problems

A researcher

Interested in topology, persistent homology, or interpretable ML applied to African data.

See the research →
03 · You want depth

A student or trainee

Mathematician, statistician, or engineer ready for short-form, graded, research-led programs.

See the catalog →
04 · You back institutions

A funder or partner

An institution that wants to fund a cohort, sponsor a research theme, or build with us.

Get in touch →
Charter

Five things we will not do.

What the institute does, on a single page.

01

Research

Mostly at the intersection of topology and applied machine learning. Persistent homology, interpretable models, asymmetric topology, a small bet on quantum-topological methods for finance. The papers are on arXiv; the code is on GitHub. Six active themes →

02

Products

Research turned into things banks and microfinance institutions can deploy. Right now: credit scoring and anomaly detection for institutions in the BCEAO zone. Later: agricultural risk. Later still: health. How we build →

03

Training

Short cohorts, taught by the people doing the research. Two days to two weeks. Every program ends with a graded defense. Bilingual delivery. Philosophy → · Catalog →

04

Incubator

AIRINA Ventures opens in year three. Not active yet. Written down on this page so we hold ourselves to it. What it will be →

Right now
Live state of the institute
Currently drafting

Persistence diagrams of microcredit default networks

Methods paper targeting Q3 2026 submission. Companion code on GitHub once accepted.

Currently teaching

TDA for Data Scientists — Cohort 1 syllabus finalization

Three days, in-person Cotonou, ~12 participants. Hands-on with GUDHI, Ripser, giotto-tda.

Currently deploying

Credit-scoring pilot · BCEAO zone

Partner under NDA. Methods will be published on GitHub at the end of the engagement.

N° 01Vol. I · 2026
Paper Club · Bi-weekly · Open

We read what we want to write.

A reading group on topological data analysis, interpretable ML, and applied mathematics for African systems. One paper, two discussants, an hour of careful argument. Read it if you can; come anyway if you can't — everyone is welcome, including first-timers. Hybrid: in-person at the Cotonou office or by Zoom.

Cycle 1 · In progress Next paper · date announced via mailing list

Get on the list.

We're working through Cycle 1's reading list now. Each session's paper and date go out on the mailing list a week ahead. Read it if you can; come anyway if you can't — first-timers welcome. Hybrid: in-person at the Cotonou office or by Zoom.

Cycle 1 Foundations to deployment
  1. 01

    Topology and Data

    Carlsson, G. (2009) · Bulletin of the AMS 46(2), 255–308 · DOI

    The canonical introduction to TDA. Every subsequent paper assumes you have read this one.

  2. 02

    Topological persistence and simplification

    Edelsbrunner, H., Letscher, D., & Zomorodian, A. (2002) · Discrete & Computational Geometry 28(4), 511–533

    The founding algorithm of persistent homology. Where the field was named and proven.

  3. 03

    On the local behavior of spaces of natural images

    Carlsson, G., Ishkhanov, T., de Silva, V., & Zomorodian, A. (2008) · International Journal of Computer Vision 76(1), 1–12

    The most beautiful applied-TDA paper: a Klein bottle hiding in the patches of every photograph.

  4. 04

    Topological methods for the analysis of high-dimensional data sets and 3D object recognition

    Singh, G., Mémoli, F., & Carlsson, G. (2007) · Eurographics Symposium on Point-Based Graphics

    Mapper, the second pillar of TDA. Required before anything Mapper-shaped shows up in our own work.

  5. 05

    Persistence images: a stable vector representation of persistent homology

    Adams, H., Emerson, T., Kirby, M., et al. (2017) · Journal of Machine Learning Research 18(8), 1–35

    The bridge from persistence diagrams to standard ML pipelines. Required before any downstream classifier reads a TDA feature.

  6. 06

    Behavior revealed in mobile phone usage predicts credit repayment

    Björkegren, D., & Grissen, D. (2020) · The World Bank Economic Review 34(3), 618–634

    The literature directly under our Phase 1 products. We agree, we disagree, we read it carefully.

Cycle 2 On deck
  1. 07

    Stability of persistence diagrams

    Cohen-Steiner, D., Edelsbrunner, H., & Harer, J. (2007) · Discrete & Computational Geometry 37(1), 103–120

    The stability theorem. Why persistent homology is a real signal and not noise.

  2. 08

    Statistical topological data analysis using persistence landscapes

    Bubenik, P. (2015) · Journal of Machine Learning Research 16, 77–102

    The other major vectorization of persistence diagrams. Bubenik vs. Adams, head to head.

  3. 09

    Stop explaining black-box machine-learning models for high-stakes decisions and use interpretable models instead

    Rudin, C. (2019) · Nature Machine Intelligence 1(5), 206–215

    The case for interpretable models, not post-hoc interpretations of black boxes. Why AIRINA products must be the former.

  4. 10

    Topology-based data analysis identifies a subgroup of breast cancers with a unique mutational profile and excellent survival

    Nicolau, M., Levine, A. J., & Carlsson, G. (2011) · PNAS 108(17), 7265–7270

    The high-stakes-medicine TDA paper. Precedent for our Phase 3 health work.

How to join Read the next paper, email paperclub@airina.africa, and show up Thursday at 16:00 WAT. No RSVP needed for in-person; Zoom link sent after email. Full Paper Club page → for format, logistics, and the archive in progress.

What's happening.

May 2026

Training catalog goes live — bilingual EN / FR

Six programs across three tiers, with two anchor cohorts opening Q3 and Q4 2026. See the catalog · version française.

Q3 2026

Cohort 1 of TDA for Data Scientists opens for registration

Three days, in-person Cotonou, ~12 participants. Hands-on with GUDHI, Ripser, giotto-tda. Full curriculum.

2025

AIRINA Labs founded in Cotonou

Phase 1 of three: applied AI for financial inclusion in West and Central Africa. Our research themes.

The plan, ten years out.

Phase 1 Years 1–3 Financial inclusion. Deep research, applied products, first training cohorts. Active
Phase 2 Years 3–5 Agriculture. Natural extension, overlapping communities, major continental funding. Planned
Phase 3 Years 5+ Health. Highest complexity, highest impact, entered at full institutional maturity. Future

Four words that have to mean something.

Rigor

We publish before we claim. We say so when we don't know.

Relevance

The questions we work on come from African institutions. Benchmarks from elsewhere do not earn a pass.

Inclusion

We exist for the people the formal economy has excluded. If our work cannot reach them, it is not the work we want to do.

Continuity

We are building an institution, not a project. The decade matters more than the quarter.

The people.

Yaé Ulrich Gaba

Founder · Director

Mathematician working at the intersection of topology, geometry, and applied AI. Research interests include topological data analysis, persistent homology, fixed-point theory, and machine learning for African systems.

Domini Jocema Leko Moutouo

Research collaborator

Mathematician and frequent research collaborator with Y. U. Gaba on topology and applied analysis.

Open positions Researchers · Trainees · Engineers — if you are a mathematician, statistician, or engineer interested in applied AI for African systems, get in touch.
Stack

What's on the bench.

For TDA
GUDHI · full persistence pipeline
Ripser · fast Vietoris–Rips
giotto-tda · TDA × scikit-learn
persim · diagrams & images
KeplerMapper · Mapper graphs
For interpretable ML
InterpretML · glassbox + blackbox
SHAP / LIME · post-hoc attribution
EBM · explainable boosting
DiCE · counterfactuals
Aequitas · fairness audit
For modeling
PyTorch · neural training
scikit-learn · classical ML
XGBoost · gradient boosting
Hugging Face · pretrained models
Qiskit / PennyLane · quantum circuits
For data & shipping
pandas / Polars · dataframes
PostgreSQL · primary store
Jupyter · exploration
FastAPI · service layer
Docker · containerization
DVC · data versioning
Languages
Python · primary
R · statistics
Julia · scientific computing
SQL · query layer

Get in touch.

HQ

Cotonou
Benin Republic

Office hours

Tuesdays 14:00–16:00 WAT
Walk-in or by Zoom