Precision Physiques
CoachingThe complete platform for a one-person online coaching business. AI drafts each client's nutrition plan, and the coach approves each one before it ships.
A one-person coaching business writes each client's nutrition plan by hand, which caps how many clients one coach can properly serve.
The plans hold special-category health data under UK GDPR, so the system had to meet that standard from day one.
Letting a model write the whole plan fails here. A plan is arithmetic first, and a wrong calorie target is wrong for a real person for a week. With the coach setting his own goal adjustments, nearly half the targets across a measured spread of realistic profiles fell below the client's own resting burn.
A marketing site, a client-accounts app and the coach console that runs both. A deterministic engine (Mifflin-St Jeor) sets each client's calorie and macro targets. AI drafts the meal plan around them, honouring allergies and preferences.
The engine owns the numbers and the model owns the food. Model output never becomes a calorie target. Targets are floored at the client's own resting burn and at a hard minimum set by sex, and no plan is generated below the very-low-calorie line. When a floor applies, the coach sees it and by how much.
An early allergy filter matched words in food names. It was hardened five times and beaten five times. In 48 plans for 8 realistic clients, 12 of the 14 failures were it refusing safe food: peanut butter for someone avoiding dairy, olive oil for someone who disliked olives, oats as gluten. Removing it took the run from 71% to 96%.
Now each client has a vetted food list, and a food off the list can't appear on their plan. The coach sees the reasons and can exclude anything. The platform won't let him allow a food whose packet declares, at high confidence, an allergen that client named.
The console: cohort roster, six plan slots per client, a wizard that regenerates a plan at a new calorie target, an editable food library, messaging, progress photos, and weigh-ins that reflow each plan a client holds.
A drafted week too far from its target is held, flagged and can't be released. A refusal costs the coach a regenerate; a bad plan would cost a client a week.
The coach approves each plan before a client sees it. Consent capture, an audit log and data export are built in.
It runs on its own isolated Firebase project, shared with nothing else MAST has built. A subject-access request is one call that gathers plans, messages, photos, requests and the coach's private notes, including drafts never released. Each export records who ran it and when.
What it actually looks like, running.
A one-person coaching brand gets the plan-writing capacity of a much bigger operation, with each plan approved by the coach first. Built and deployed, ready to onboard, with data protection designed in from the start.
Testing runs in two gates. The free lane uses stubs and emulators, which can't see served copy, real timing or real data. A second gate runs the real thing, model calls included, before anything ships, because a green emulator suite once hid four faults on the deployed site.
The coach's judgement sits on top, the engine does the arithmetic, and client data was treated as special-category from the first commit.
Take half an hour to see what a compliance-grade client platform could look like for your business.
No commitment and no fixed quote until it is scoped. If we go further, you own what we build.