People stay in charge of every AI system we build
We are a new company, so these are commitments rather than a track record. On every project a named person reviews what the AI produces, your data stays where we agreed, and each system is documented, including where it can fail.
[Certification]- [Certification]
- [Certification]
- [Certification]
- [Certification]
A badge appears here only once a certification is held and current.
Three commitments on every project
Human review
AI drafts, sorts and suggests. A named person checks anything that affects a citizen, a patient, a customer or a student before it is acted on.
Privacy by design
We collect only what the project needs, agree in writing where it is stored, and never use one client’s data to build tools for another.
Secure deployment
Access controls, logging and an incident plan are agreed before launch, whether the system runs on your own servers or on a cloud you approve.
One record for every AI system you run
On every project we keep a short register: who owns each system, what data it touches, how it was tested, and who to call when something goes wrong. It stays with you after handover.
AMSAI use register
ExampleEvery AI system in use, its owner, its data and its review status.
All systems
Under review
Approved
Retired
Search by system or owner
ApprovedReport drafting assistant
Drafts monthly programme reports for staff to check and sign off.
Owner: [Name]
Under reviewSomali FAQ assistant
Answers routine citizen questions and hands the rest to staff.
Owner: [Name]
ApprovedDocument summariser
Summarises field reports; every figure is checked by the M&E team.
Owner: [Name]
PilotRecords search
Finds files across scanned archives for authorised staff only.
Owner: [Name]
What this looks like in practice
Fairness testing in Somali and English
- We test outputs on real Somali and English text before anyone relies on them
- We check results across the groups a system serves, including regions and dialects
- We report what we find, including the gaps we could not close
SO
Somali test set
EN
English test set
Evaluation run
EXAMPLE
Somali accuracy
[result]
English accuracy
[result]
Dialect coverage
[result]
Known gaps
[listed]
Reviewed by a person
System card
ONE PAGE · EVERY SYSTEM
Purpose
Data used
Known limits
Human reviewer
Owner
“You are talking to AI”
Transparency and documentation
- People are told when they are dealing with AI
- Every system comes with a one-page record: purpose, data, known limits and owner
- Documentation is handed over in Somali and English
Data handling
- We use only the data a project needs
- Where data is stored is agreed in writing before work starts
- Client data is never used to build tools for another client
Your data
Records · documents · conversations
Stays in: agreed location
ENCRYPTEDACCESS-LOGGEDincident-plan · example
1
Report received
WhatsApp, email or staff
2
Named contact responds
within the agreed time
3
System paused if needed
owner decides
4
Review with you
what changed and why
Incident response
- A named contact and an escalation path are agreed before launch
- Logs are kept so problems can be traced and fixed
- Every incident is reviewed with you afterwards
Raise a concern
If an output from one of our AI systems looks wrong, unfair or unsafe, tell us. We will look into it.
Every report goes to a named person on our team. We reply, and we tell you what we changed.
Email [email protected] or message us on WhatsApp +252 61 528 4463. How we handle reports is set out in our Responsible AI policy.
Writing on responsible AI
[Published article]
Shown once an article is published, with a named author.
[Published article]
Shown once an article is published, with a named author.
[Published article]
Shown once an article is published, with a named author.
