AboutA research collective, not a lab.
GCT Exchange is a distributed, non-profit network of 37 researchers, engineers, clinicians, agronomists, and linguists working across nine African countries. We pool compute, mentorship, and — most importantly — the patience to do applied research properly.
Why we exist.
Most of the AI systems that touch African lives today were not built in Africa, not evaluated on African contexts, and not maintainable by African teams. The gap is widening: frontier models concentrate in a handful of labs, training data under-represents African languages by an order of magnitude, and the deployment patterns that work in Nairobi or Lagos rarely look like the ones validated in Silicon Valley.
GCT Exchange exists to close that gap from the inside. We fund African researchers to work on problems their communities actually have. We release everything we build under permissive licenses so others can verify, extend, and challenge it. And we measure ourselves against deployment in the field, not paper acceptance rates.
We are deliberately small for our ambition. Thirty-seven people, five program leads, and a very small operations team. We would rather ship one useful model than five papers nobody reads.
How we work.
Six principles that show up in every grant proposal, every hire, and every release we ship.
African-led
Every research question we work on is named, scoped, and prioritised by African researchers in conversation with the communities the work serves. Funders fund; they do not direct. Our published pillars are decided by the network and revisited every two years.
Open by default
Datasets, model weights, evaluation code, training recipes, and reproduction scripts are released openly — typically CC-BY or CC-BY-SA for data, Apache-2.0 for code and weights. The bar for keeping something closed is high and documented in writing for each exception.
Useful before novel
We optimise for work that runs in a clinic, on a farm, in an extension office, or on a $40 phone — not for the marginal arXiv citation. Where the two align, wonderful. Where they don't, the deployment wins.
Slow where it matters
Consent processes, language work, clinical validation, and field co-design take real time. We build that time into our schedules, our grant proposals, and our publication expectations. We will miss a deadline before we will rush a community review.
Honest about failure
Every annual report contains a 'what we got wrong' section. We publish negative results, retracted approaches, and budget overruns. The credibility of open research depends on it.
Pay people properly
Salary bands are public, benchmarked to local cost of living, and audited yearly. We back-pay when we get it wrong. We do not run unpaid internships.
Timeline.
A short history. The longer one is in the annual report.
Twelve researchers from six countries write the first version of the principles document.
Eight fellows across Languages and Agriculture. Six are still in the network.
AfriParallel-NG, the Nigerian language-pair parallel corpus, released under CC-BY-SA 4.0.
Nairobi hub launches with Dr. Lerato Ndlovu as founding director.
SahelClim-30 dataset begins collection in collaboration with AGRHYMET.
Audited financials and 'what we got wrong' section published publicly.
First open Kinyarwanda speech recognizer that runs usefully on low-end Android.
Leadership.
The people accountable for our programs and our operations.
Kigali · PhD, University of Edinburgh. Leads the Kinyarwanda program and chairs the board.
Accra · Joint appointment with KNUST. Twenty years on smallholder advisory tooling.
Nairobi · Public health physician. Previously WHO AFRO and Kemri-Wellcome.
Bamako · Earth observation lead. Long-running collaboration with AGRHYMET.
Lagos · Builds the reproducible-research stack. Previously an early engineer at Andela.
Dakar · Runs fellowships, residencies, and the open-call microgrants.
Accra · On-device ML and quantization. Co-lead on the cassava field stack.
Ibadan · Chairs the project ethics committee. Non-voting on the board.
Read the annual report
Outputs, financials, demographics, and a candid section on what we got wrong in 2025. Forty-eight pages.
Download · annual-report-2025.pdfWork with us
We hire researchers, engineers, clinicians, and program staff across the continent. All bands published.
See open roles →