Adaptive AI trainer for a coaching practice
A coach-facing loop that plans the week from a structured client state, logs the session, and recalculates load — without inventing a medical clearance.
What changed. One coach can keep individual weekly plans current for a roster of more than a hundred clients, with pain and missed sessions treated as inputs rather than ignored.
Stack. structured client state · weekly plan generation · load adaptation · coach dashboard
Context
A coach with a growing roster could not write a fresh programme, answer the day’s questions, and adjust load for every client by hand. Club apps offered video libraries and a static calendar. Copy-paste templates ignored sleep, soreness, and the session that did not happen. Clients stalled; the coach’s evenings disappeared into spreadsheets.
Task
Give the coach a loop that drafts an individual week from a structured state file, records what was actually done, and recalculates the next week. The system must scale the coach’s attention, not replace the coach. Pain, missed sessions, and “I slept badly” are first-class inputs. The loop must not talk past an injury as if it were a motivation problem.
What we built
Each client is a state.json: current level, fatigue, sleep, joint notes, and history. The engine reads that file and writes a week that combines a daily joint-mobility sequence with progressive bodyweight strength — six primary moves, each with a ten-step ladder. Intensity on the mobility block follows yesterday’s load and this morning’s flags. The strength block picks the step, sets, and reps from the recorded ladder, not from a slogan.
The client logs reps and how the set felt. State updates. The next week is regenerated from the new file, not from a calendar that pretends the missed Tuesday still counts. The coach sees a roster summary on one dashboard instead of opening a chat per person. Programme drafting that used to take about forty minutes per client now takes about two.
When the state records pain in a joint, the loop is supposed to regress that movement or escalate to the coach — not “adapt around it” with a harder variation.
What changed
One coach can keep weekly plans current for more than a hundred clients at once. Clients get a revised week when life interrupts, without waiting for office hours. The coach spends time on the cases that need a human, not on retyping the same ladder.
What we would do differently
Make injury and pain a hard refusal in the harness, not a paragraph in the prompt. The first version was too willing to keep the week “moving” when the honest output was: stop this pattern, flag the coach, do not invent a workaround. Adaptation is a privilege of a clean state file.
Related service: AI consultant and chatbots