Skip to content
MALCOMv2.5.0, home

CoachMauriceKE

AI-assisted fitness coaching portal delivered as an installable Progressive Web App.

CLIENT / VENTURE
CoachMauriceKE
ROLE
Full-Stack Developer
TECHNOLOGIES
Vite, Tailwind, Gemini, Groq
DELIVERY
In Production

The Problem

Personalised meal plans were written by hand for every client, which capped how many people the coach could serve. Publishing a native app would have meant high app store fees and a slow release cycle, and a single AI provider made plan generation fragile whenever that provider was slow or down.

What I Built

I built a client portal as a Progressive Web App that installs straight from the browser. Clients enter their goals and preferences, and the system generates a nutritional matrix and meal plan automatically. Generation runs through an LLM fallback chain, so a failed request to one provider is retried on the next without the client noticing.

Core Architecture

  • Vite & Tailwind PWA Frontend

    Lightweight installable client with offline caching, so programmes stay available on weak mobile connections.

  • LLM Fallback Chain

    Gemini as the primary model with Groq as a fallback, behind one interface that validates every generated plan.

Key Challenges

  1. Uninterrupted Generation

    Detected provider timeouts and malformed responses early and retried on the fallback model, keeping plan generation reliable.

  2. App Store Independence

    Delivered a native-feeling install and offline experience without app store distribution or its revenue cut.