Performance Command Center
AI-assisted training recommendations that coaches stay in charge of
Every athlete receives a predictable Standard Plan. The Recommendation service reads the evidence behind that plan and, only when the evidence supports a material change, proposes one structured adjustment for a coach to accept, edit or reject.
Staging · synthetic athletes only
How a Recommendation is made
- Standard PlanThe existing planning system builds the deterministic plan as it does today.
- Evidence gatesAccess, data freshness and source coverage are checked. Missing or stale evidence stops here.
- AI proposalThe AI agent proposes one change with an explanation and the evidence it relied on.
- Safety validationDeterministic code checks every field, safety limit and the whole day and week of training.
- Coach decisionA coach accepts, edits or rejects it. Without a Recommendation the Standard Plan proceeds.
No Recommendation is a normal, good outcome. The service never invents a change to look useful.
Safeguards
Coaches decide
Every AI change needs coach approval. The service cannot approve or publish a plan.
Traceable
Each Recommendation records the evidence, model, prompt, policy and methodology versions that produced it.
Fails safe
Kill switches and rollback return every athlete to their Standard Plan without interrupting training delivery.
Progress
Built
- End-to-end Recommendation pipeline with evidence and safety gates
- AI agent running on Amazon Bedrock AgentCore in staging
- Release controls, kill switches and rollback
- Audit traces for every run
Next
- Evaluation suites that gate every new AI release
- Shadow mode: real inputs, no visible Recommendations
- A coach-visible pilot after approval