A team of 35 specialised AI experts

How a request flows

Example: a learner asks “Teach me RAG”.

Rules every expert follows

1. Least privilege — each expert only has the tools it needs.
2. Evidence before assertion — answers cite the course; missing evidence is stated.
3. Course context first — official material outranks general knowledge.
4. Human approval — publishing, grades, credentials and official records need a person.

AgentRequest {                          AgentResponse {
  learner_id, session_id, course_id,      result,
  task, context, learner_state,           evidence[ source, relevance ],
  evidence_required,                      confidence,
  allowed_tools,                          actions_taken, recommendations,
  output_format                           next_agent, safety_status
}                                       }

   DRAFT ──▶ TEST ──(Evaluation Lab ≥ threshold)──▶ APPROVED ──(human)──▶ PRODUCTION

   Model tiers:  large  → orchestration, research, course creation, QA
                 medium → tutoring, RAG answers, assessment
                 small  → routing, profiles, notifications, records
          (Gemma 4 via Ollama · vLLM · Claude · offline engine)