From AI policy
to evaluated deployment.
This is where CentPol is going. We are building toward an applied AI lab and evaluation commons — a place where institutions can evaluate models, run pilots, publish benchmarks, train talent, and govern AI for real-world use across institutional contexts.
The bottleneck has moved from access to deployment
In the next phase of AI adoption, institutions won't win by accessing powerful models — most already can. They'll win by knowing which model works for which workflow, with which data, at what cost, under which governance regime, and how to prove it works.
Global benchmarks don't speak local
AI claims are everywhere; credible, context-specific evidence is scarce. Global benchmark suites rarely reflect local languages, infrastructure, regulation, and institutional realities — especially across African public-sector, SME, education, agriculture, and civic workflows. That gap is where CentPol intends to be useful.
A real base to build the lab on
The direction below is ambitious and deliberately forward-looking. It rests on work that already exists at CentPol — not a blank page.
Policy & proposals
National and institutional technology-strategy frameworks, led by the Uganda AI workforce thesis.
Research & insights
Evidence reviews, signals, and analysis that build public credibility and institutional trust.
Training & cohorts
Structured courses, curricula, and cohort delivery — the machinery that runs real learning programs.
Fellowship pipeline
The talent engine that will, over time, produce the lab's benchmarks, datasets, and pilot work.
An applied AI lab & evaluation commons
Not a frontier model lab — that race belongs to those with vast compute. CentPol intends to compete downstream: adapting, evaluating, deploying, and governing AI in the contexts large labs are unlikely to prioritize. Five capabilities, built deliberately, each producing reusable public artifacts.
AI readiness & workflow discovery
Map institutional workflows, data, risk, and ROI — turning ambition into a concrete use-case portfolio.
Model evaluation & benchmarks
Compare open and commercial models on local tasks and risk scenarios. Publish benchmark reports, model cards, and reproducible eval scripts.
Model adaptation & deployment
RAG, agents, and selective fine-tuning — chosen only when the benchmark proves they beat simpler approaches.
Governance & safety
Risk registers, red-teaming, human validation, and data controls — the assurance regulated and public partners require.
Talent & fellowship pipeline
Fellows trained on real lab projects, producing portfolios, benchmarks, and partner-matched skills.
Compounding, not billable hours
Every engagement should leave a reusable trace — a benchmark, dataset, model card, deployment template, governance pattern, case study, or trained person. That is how the lab compounds over time.
Honest about the journey
We're sharing direction, not delivery dates. Here is how we think the work sequences — and we'll move items forward as they ship, in the open.
Policy intelligence, research, training, and the fellowship engine that the lab will run on. All live today.
An evaluation rubric and a first benchmark track; fellows producing real, documented work; governance templates taking shape.
Public benchmark reports, model cards, deployment case studies, and a living leaderboard for African-context AI — built with partners, published openly.
This gets built with partners
The fastest way to make this real is together. If any of these sound like you, we'd like to talk — early collaborators shape the agenda.
Run an AI readiness sprint
Map public-service workflows, evaluate options, and pilot safely — with governance built in from day one.
Co-create an applied fellowship
Students work on real partner evaluations and benchmarks, building employable, evidence-based AI skills.
Sponsor a local evaluation lab
Turn compute credits into credible, public local-context evaluations and responsible deployment stories.
Pilot with benchmarked ROI
A focused applied-AI pilot with measured results and governance controls — not a slideware demo.
Back inclusive, measurable AI
Workforce, agriculture, education, and civic AI — evaluated, documented, and tied to real outcomes.
Or join as a fellow
The fellowship pipeline is the engine. If you want to build the artifacts that make the lab real, start here.
Help us build the policy-to-production lab.
Bring a workflow, a benchmark idea, a cohort, or a partnership. We'll build the evidence together.