Services · Sovereign AI
Sovereign AI & Public Data Systems
National AI and data capability built with the institutions that will own and operate it.
African governments need usable AI systems, governed data, capability in the languages public services actually run on, and teams that can operate what vendors deliver. Many imported systems create continued dependency and never learn the language of the citizen in front of the counter. Kilwa Sovereign AI is designed around the opposite premise: the partner should end the engagement able to run and extend the system without us.
- 55
- member states covered by the African Union's Continental AI Strategy, which calls for coordinated national approaches, responsible governance and Africa-centred capability
- African Union, Continental AI Strategy
- US$10bn
- that a 2026 initiative plans to mobilise for African AI foundations, infrastructure, entrepreneurship and technical support: evidence of demand, not a market size
- UNDP Nairobi AI Forum, 2026
- 85
- languages in the Kilwa source registry, and small language model work in Kinyarwanda delivered with local teams
- Kilwa source registry; Kilwa Sovereign AI
- Rwanda
- the reference and product-development market, with the Ministry of ICT and Innovation, RISA and the Centre for the Fourth Industrial Revolution as programme and policy partners
- Kilwa partner record
The decision you are facing
What the committee actually asks.
- 01Which public decisions would a governed data and AI capability actually improve, and which are better left alone?
- 02Where does our data sit, who may use it, and under what residency and access rules?
- 03Can the system work in the national languages, not only in English or French?
- 04Who runs it in year three, and on what terms?
- 05How do we keep control of models, data and vendors as the capability grows?
What we deliver
The offerings, grouped by the question they answer.
Strategy and governance
- National AI and data-readiness assessment
- Sovereign AI strategy
- Data governance, residency and access design
- Responsible AI and model-risk controls
Build
- Local-language small language models
- Public-service AI pilots
- Public-sector knowledge systems
- Cloud and compute architecture
- Investment promotion intelligence systems
Transfer
- Local team training and capability transfer
- Ongoing sovereign-controlled support
The model
Capability, not dependency.
Most AI delivered to governments is rented. Kilwa Sovereign AI is designed around the opposite premise: the partner should end the engagement able to run and extend the system without us.
How Kilwa answers it differently
Not another country report.
Built to be owned
Systems are delivered to be owned, operated and extended by the partner government. The measure of success is that the partner no longer needs Kilwa for day-to-day operation.
National languages, natively
Small language model work in national languages, developed with local institutions and local data rather than adapted from models that never saw the language, with Rwanda as the reference and product-development market.
Delivered with institutions, not around them
Sovereign work is co-branded and co-delivered with the ministry, the national agencies and in-market partners, and it is only taken on where a signed local consortium, a narrow use case, clear data rights and funded capacity transfer exist.
How we engage
3 formats. One path.
Scoped to the mandate rather than a tier list, so the fee follows the decision, the markets and the horizon.
Readiness assessment and strategy
The national systems, data estate, language requirements and use cases mapped with the ministry or agency that will own the result.
Pilot and build
A narrow public-service or investment-promotion use case, built with local institutions embedded in the delivery team from day one.
Transfer and sustain
Local teams trained to operate and extend the system, with documentation and model cards handed over in full and support on terms that keep the system under sovereign control.
1Scope
The decision, the mandate, the markets and the horizon, agreed in writing before any evidence is assembled.
2Evidence
Inputs gathered from the source registry and the client, each flagged verified or estimated, with provenance recorded.
3Read
Structure and timing read separately on the published gauges, with the factors behind each shown.
4Decide
A decision pack for the committee, board or ministry: the call, the confidence, the trail and the condition that would prove it wrong.
5Monitor
Assumptions become dated signposts, watched between editions and graded in public, misses included.
Named solutions
Delivered inside the engagement.
Four products serve four points in the same decision cycle. This service delivers one of them.
- Discover and compareKilwa Africa Factbook
- MonitorZawadi Briefing Engine
- Underwrite and approveKilwa Deal Room
- Prove what happenedKilwa ImpactOS
Kilwa ImpactOS
Know what changed. Prove why. Fund what works.
The governed evidence infrastructure behind public-sector decisions: capital, programmes, outcomes and evidence in one permissioned data model, with local-language access and interoperability with the systems governments already run.
Partners
Delivered with institutions, not around them.
Sovereign work is co-branded and co-delivered. Partner institutions appear alongside Kilwa at equal prominence, because the capability is meant to be theirs.
Partners
Formal agreements and programme memberships — the research, language, data and infrastructure relationships the platform is built on.
Northwestern UniversityExperiential Learning Agreement (March 2026) and NSF SBIR/STTR Phase I research support
Google CloudGoogle Cloud Partner Advantage member
Google for StartupsStartup programme member
Low-resource language model partner, Hong Kong
Kinyarwanda language data and model partner, Rwanda
Engineering partner, Rwanda — sovereign AI and national e-procurement
Data platform partner
Voice and speech AI partner
Collaborators
The institutions Kilwa builds and delivers with on the ground, principally across Rwanda's sovereign AI programme.
Boundaries
What this service is not.
- A selective, consortium-led service, not a general AI consultancy: engagements need a signed local consortium, a narrow use case, clear data rights, funded capacity transfer and realistic infrastructure requirements.
- Restricted government data never enters shared training, embeddings, caches or public retrieval.
- Kilwa's platform scores are structured risk rankings, not validated predictive models, and are never repackaged as a sovereign's own statistics.
Kilwa Technologies LLC is not a registered investment adviser, broker-dealer, credit rating agency or law firm. Its services are market, policy and country intelligence and decision support, provided under a written engagement. They are not investment, legal, tax or accounting advice, and nothing on this site is an offer or solicitation.
Before the first call
The questions buyers ask.
- What makes AI sovereign?
- Ownership. The models, the data and the operating capability sit with the partner government, in national languages, with local teams trained to run and extend them. Capability that arrives without ownership becomes a dependency.
- Where has this been done?
- Rwanda is the reference and product-development market, with the Ministry of ICT and Innovation, RISA and the Centre for the Fourth Industrial Revolution as programme and policy partners, and small language model work in Kinyarwanda delivered with local teams.
- What does Kilwa need before taking on a sovereign AI engagement?
- A signed local consortium, a narrow use case, clear data rights, funded capacity transfer and realistic infrastructure requirements. Without those, the honest answer is no.
- Does government data train Kilwa's models?
- No. Restricted government data never enters shared training, embeddings, caches or public retrieval.
Research for this service
All insights
Built for: Governments & Public Institutions. Other services: Capital allocation & deals · Risk & regulatory intelligence · Market entry & growth · Sovereign capital & investment promotion.
Scope an engagement
Tell us the decision. We will tell you what evidence it needs.
One conversation to agree the mandate, the markets and the horizon. A reply within one business day, from a person.