AI Programme Personalisation Engine
AI Programme Personalisation Engine
Available from v1.0.89
The AI Programme Personalisation Engine is the core intelligence layer of the platform. It analyses each member's training history and physiological readiness to produce a fully personalised weekly training programme — blending tracks, adjusting loads, and surfacing coaching context generated by Anthropic Claude.
How It Works
Each week, the engine runs a pre-computation pipeline (via Inngest background jobs) that processes per-member data across four dimensions:
| Input | Description |
|---|---|
| Session Logs | Historical attendance, completed work, and volume across all tracks |
| PR History | Personal record trends used to gauge strength and fitness trajectory |
| Readiness Scores | Current recovery and readiness indicators (e.g. sleep, HRV, subjective feel) |
| Member Goals | Stated training goals (e.g. build strength, improve conditioning, lose weight) |
These inputs are combined to produce a personalised weekly plan that:
- Blends training tracks — Weights are assigned to each track (strength, conditioning, gymnastics, etc.) based on the member's goals and programme template set by their coach.
- Adjusts loads — Prescribed loads are scaled up or down relative to the member's recent performance and readiness, reducing injury risk and preventing overtraining.
- Generates coaching context — Anthropic Claude produces a narrative rationale for the week's prescription, giving both coaches and members a clear explanation of why the programme is structured the way it is.
Pre-Computation with Inngest
Personalised plans are computed asynchronously ahead of time — members never wait for their programme to be generated in real time.
- Inngest jobs are triggered on a weekly schedule.
- Each job processes a single member's data and writes a ready-to-deliver plan.
- Coaches can review AI-generated plans before they surface to members (depending on gym configuration).
- If a member's data changes materially mid-week (e.g. a significant readiness drop is logged), the engine can re-compute an updated plan.
Coaching Context (Claude)
For each weekly plan, Claude receives a structured prompt containing:
- The member's recent training load and fatigue indicators
- PR trends and areas of strength or weakness
- Current readiness score
- Stated goals and any coach notes
Claude returns a short coaching narrative — a plain-language explanation of the week's focus, the reasoning behind load selections, and any cues the member should keep in mind. This context appears alongside the programme in the member's daily view and the coach's dashboard.
Track Blending
The engine supports multi-track programming. Rather than assigning a member to a single track, it can blend prescriptions from multiple tracks in proportions derived from:
- Member goals — A member focused on strength will receive a higher proportion of strength-track work.
- Coach-configured templates — Gym owners and coaches define base programme templates that constrain or guide blending ratios.
- Historical adherence — Tracks the member consistently completes are weighted more favourably.
Load Adjustment
Prescribed loads (weights, rep ranges, intensity percentages) are adjusted from baseline programme values based on:
- Readiness score — Low readiness results in reduced load prescriptions to prioritise recovery.
- Recent PR history — Upward PR trends allow the engine to safely increase load targets over time.
- Session log volume — Unusually high recent volume triggers a conservative load week automatically.
Roles & Access
| Role | Capability |
|---|---|
| Gym Owner | Configure programme templates; view personalisation activity across all members |
| Coach | Review and edit AI-generated plans before delivery; see Claude coaching context |
| Member | View their personalised weekly plan and the coaching rationale in their daily experience |