An evidence-first view for families, schools, and partners. Understand how content is checked, when humans step in, and how decisions are recorded—not legal advice, but how the platform actually works.
Administrators can open the live dashboard directly. If you are not signed in, you will be prompted to sign in—then use Admin → AI regulation in the menu. Everyone can read the sections on this page without an account.
Content prompts are checked against configurable rules by grade and education level before generation runs. Blocked requests never reach the lesson or curriculum builder.
Borderline or flagged content can be queued for review. Lessons stay unpublished until an approved administrator clears the queue when policy requires it.
Each stage logs outcomes with timestamps, prompt hashes (not raw text in exports), model identifiers, and links to resources. CSV and JSON exports support governance reporting.
These diagrams summarise the same pipeline and decisions recorded in the administrator dashboard. They are illustrative, not a legal guarantee.
From request to learner: each step can write an audit event. Human review happens before publish when the system flags an item.
A strict block stops the request. A “near miss” may still allow generation but queues review—depending on your admin settings.
Administrators set a lower and upper percentage. Marks in the middle band are flagged for oversight while students still receive feedback.
Active and enforced. Guardrail rules are versioned in application settings; regulation thresholds (for example review bands) can be adjusted by administrators within safe defaults.
Allow · Block · Review required. The system records how often content is allowed straight through, blocked at policy, or held for human review—so trends are visible over time.
“Near miss” keyword patterns, unusual length, or AI grading confidence in a review band can raise a flag. That does not always mean something is wrong—it means a person should verify.
Built for parents, teachers, school leaders, and partners who need a shared vocabulary: pipeline stages, outcomes, and evidence—not hidden inside engineering-only tools.
Plain language for the same labels you will see in technical reports and exports.
Passed automated checks for that stage. Content may still be draft until a teacher or admin publishes it.
Automated checks suggest human verification—e.g. soft guardrail signals or grading confidence in a review band.
Policy does not permit generation or publication for that input. The user sees a clear refusal, and the event is logged.
How Mwalimu aligns intent from policy settings to learner-facing experiences.
Mwalimu is designed for CBC-aligned teaching and learning, with local examples in AI prompts where appropriate. Data handling should follow the Data Protection Act, 2019 and your institution’s policies. National AI and digital policy continues to evolve—including the Kenya AI Strategy 2025–2030 direction—so we keep the architecture modular for future regulatory sandboxes and notices.
This page describes product behaviour, not legal advice. Consult your lawyer and, where relevant, the Office of the Data Protection Commissioner (ODPC) for compliance questions.
Kenya’s direction—including the Kenya AI Strategy 2025–2030 and expected evolution of law for AI and emerging technology—emphasises responsible innovation, data protection, human oversight, and trust. Mwalimu is designed to support those goals in EdTech through built-in controls and evidence, not through a government “seal of approval” on the product itself.
Clear responsibility for AI-assisted decisions, documentation, and oversight.
Regulation events (append-only), admin dashboard, exports (CSV/JSON), and review queues so schools can show who approved what and when.
Personal data must be processed lawfully, fairly, and securely (DPA 2019).
Role-based access, no raw prompts in default exports, hashes for audit, and alignment with your institution’s privacy notices and ODPC expectations.
High-risk or sensitive use of AI should support human review where appropriate.
Guardrails block disallowed requests; “near-miss” and output checks can queue content for admin review; lessons can stay unpublished until review; AI marks in an “uncertain” band are flagged for oversight.
AI in education should respect curriculum, language, and local relevance.
CBC-oriented lesson generation, Kenyan examples in model instructions, admin-editable guardrail rules by grade/level, and short transparency text for teachers on AI-generated or AI-marked content.
Users should understand when AI is used and on what basis.
Public page (this site), policy intelligence copy for families, and rubric-style explanations on assessments where the model grades free text.
Regulatory approaches may include sandboxes, standards, and new instruments.
Modular architecture (Next.js + optional coursegen workers), versioned settings, and internal webhooks so controls can adapt as rules and sector guidance are updated—without replacing your legal counsel or ODPC process.
Important
Alignment with national policy is implemented as operational and technical measures in the product. It does not replace registration with the ODPC, legal review of your processing, or future licensing if the law requires it. We recommend periodic review with your lawyer and the latest official publications from the Ministry of Information, Communications & The Digital Economy and the broader digital economy framework.
Official sources for data protection, ICT policy, and curriculum—not operated by Mwalimu, but relevant when discussing AI in schools.