Case study

Hā Health: Comprehensive AI Wellness Ecosystem

Hā Health wanted one AI-driven wellness experience: personalised meal, exercise and meditation recommendations, recipes written on demand, and meal plans that turn into a shopping list, on web, iOS and Android.

Client: Hā Health
Industry: Health & Fitness
AI capability: Generative AI & Recommender Engines

Visit Hā Health

Three phones showing the Hā app: an exercise screen, AI meal recommendations and meditation recommendations

Platforms: web, iOS and Android: 3
AI modules in one ecosystem: 5
On the App Store and Google Play: Live
Client location: Australia

The challenge

Personal wellness advice has to be safe, instant and the same everywhere

An AI wellness platform is more than a recommendation feed. Recipes and meal plans have to respect strict dietary restrictions and medical needs, so model output must be controlled: no hallucinated ingredients, no unrealistic quantities. At the same time, recommendations for meals, exercise and meditation had to be generated instantly for a growing user base.

The platform also had to build meal plans, generate shopping lists, and compare ingredient prices across retailers, then deliver all of it identically on the web app, Android and iOS.

Meals people can actually eat
Plans and recipes that respect health goals and dietary restrictions at scale.
Routines that keep up
Exercise and meditation guidance that adapts as users log progress.
From plan to basket
Shopping lists tied to the plan, with live grocery prices and availability.
One experience, three platforms
The same AI behaviour on web, iOS and Android.

What we built

Five AI modules behind one wellness app

Instead of one large model doing everything, we split the product into specialised engines. Each one is small enough to be fast and specific enough to be checked.

Stage 1
Capture the user's profile
Health goals, dietary restrictions, fitness level, preferred workout types, equipment and available time are collected as structured preferences, not free text.
Structured user profile
Stage 2
Recommend meals
A custom lightweight recommender suggests meals from health goals and preferences, with dietary restrictions applied before anything is shown.
Meal suggestionsRecommender engine
Stage 3
Recommend exercise and meditation
An adaptive exercise engine builds routines for the user's fitness level and adjusts as they log progress. A meditation engine suggests sessions based on mood, stress level and the time available.
Routines and sessionsAdaptive recommendersFeedback loops
Stage 4
Write new recipes inside guardrails
An agentic workflow collects precise health and dietary constraints from the user, then writes recipes with anti-hallucination guardrails: ingredients, quantities and steps are checked before the recipe is shown.
Nutrient-aware recipesAgentic workflowAnti-hallucination guardrails
Stage 5
Generate the plan and the shopping list
The meal plan generator combines recipes into weekly plans, consolidates the ingredients into a shopping list and compares prices across major retailers.
Weekly plan with priced grocery listPrice comparisonReal-time availability

Wellness plan · sample user

Personalising

Sample user

Goal: eat better and manage stress

Diet

Vegetarian
High protein
Low FODMAP

Workouts

Yoga
Walking
No equipment

Mindfulness

Stress relief
10 min a day

Sample weekly plan
Day	Meal	Move	Mind
Mon	Chickpea scramble	Yoga 20′	Calm 10′
Tue	Lentil tacos	Walk 30′	Sleep 10′
Wed	Tofu stir-fry	Yoga 20′	Focus 10′
Thu	Quinoa bowl	Walk 30′	Calm 10′
Fri	Veggie curry	Stretch 15′	Breath 10′

5 days planned from 3 recommenders. Users can also build a plan themselves.

Shopping list5 items

Sample shopping list with prices at two stores
Item	Store A	Store B
Spinach · 2 bags	$5.00	$4.40
Capsicum · 3	$3.60	$3.90
Chickpeas · 3 cans	$3.30	$2.70
Quinoa · 500 g	$6.50	$5.90
Firm tofu · 2 packs	$7.00	$7.40
Total	$25.40	$24.30

Cheapest mix: $23.60WebiOS · Android

1/4The user shares goals and preferences: diet, workout style, equipment and how much time they have.

Illustrative sample: the profile, recipes, stores and prices are made up for this walkthrough.

Results and impact

Complex health management, reduced to a plan and a list

The modular setup made personalised wellness simple for Hā Health users: the app suggests what to eat, how to move and how to wind down, then turns the week into a shopping list.

Personalised plans generated in the app, not assembled by hand
Shopping automated, with prices compared across retailers
Better engagement with and adherence to personal health routines

We build the same kind of guarded generative workflows for other wellness products, including the gbMeals AI nutritionist, and apply similar controls to clinical LLMs. See the rest of our healthcare AI work.

Specialised engines
Each recommender does one job, which keeps responses fast and easy to test.
Guardrails on generation
Recipes are checked against dietary restrictions and realistic quantities before users see them.
Guardrails for health LLMs
Shopping handled
Weekly plans become a consolidated list with prices compared across retailers.
Cross-platform delivery
The web app and the iOS and Android apps all call the same AI services.