AI for sport
The AI prepares. The coach decides.
We don't promise to predict injuries. We show you the load that causes them, today, per athlete, with the maths in plain sight. That is what a coach can act on tomorrow morning.
Why we don't claim to predict injuries
The published state of the art in injury prediction, from teams working with first-division clubs, sits around 33% precision. Two in every three alerts are false. A coach who gets three warnings a week and is right about one stops looking at warnings within a month.
What is actually true
Acute-to-chronic load, sleep, soreness and velocity loss don't predict the future: they describe the present. And the present is actionable, you can drop a set, postpone a test, move a day.
What we do with it
We compute it, we show the computation, and we leave the decision with the person accountable for it. No number appears that you can't open up and trace.
Why this wins
A closed model that is right a third of the time asks for blind trust. An explainable metric asks for judgement, and the coach's judgement is already there.
Real case
PeakRaptor: scientific training, built on this architecture
A training platform for high-performance athletes and coaches, in production. Not a demo: software coaches use to plan weeks and athletes use in the gym, set by set.
Load and readiness
Acute-to-chronic workload ratio, a daily 0-100 readiness score and a wellness questionnaire (fatigue, sleep, soreness, stress, mood). A heatmap puts the whole squad on one screen.
Strength measured, not declared
1RM is never typed in: it comes from every logged set, weighted by confidence (velocity > RPE > reps) and decayed over time. PRs are detected mid-session.
Velocity as a signal
VBT data tied to the actual session. Velocity loss within a set is the fatigue marker perception misses.
Real periodisation
Mesocycles, microcycles, non-uniform per-set prescriptions, and competitions that pull the taper. The AI drafts the plan; the coach approves or rewrites it.
The subjective data belongs to the athlete
RPE and the wellness questionnaire can only be filled in by the athlete, never by the coach on their behalf. It looks like a permissions detail; it is what separates a number you can trust from one typed in a hurry at the end of the day.
An assistant with the full history
An assistant that answers from the athlete's REAL history, the coach's protocols and the platform's patterns, not from internet generalities.
What the architecture adds
What sets this apart from a pretty dashboard isn't the chart. It's what sits underneath.
Every action is attributable
Who changed what, when, and what it touched. An append-only trail, not a log you can erase.
The AI proposes, never acts alone
Anything that mutates data goes through a decision queue. A proposal can be wrong; an action executed silently can't be undone.
The data belongs to whoever generates it
Athlete data is near-health data, often from minors. Per-customer isolation, and we don't train models on it.
Open to agents, on a leash
The same capabilities exposed over MCP and OAuth, with per-application scopes, so you integrate with what you already have without migrating anything.
Who this is for
Coaches and academies
You want the platform. PeakRaptor is in production and takes athletes today.
Clubs, federations and sports companies
Your data is spread across spreadsheets, wearables and an ERP. Here the conversation is about what connects to what, and what the AI can prepare for your people to decide.
Talk to us
Tell us where you're coming from. We reply within 24 hours, and the first reply is from a person.