Leading Consumer Healthcare Brand - A Gamified Kids' App Without an 18-Month ML Build
How we redesigned a children's wound-care companion app so it could launch without custom computer vision or live AI calls, while keeping the full gamified experience.
The numbers, at launch.
A leading consumer healthcare brand wanted to make childhood injuries less frightening, with a digital companion app that turns a scraped knee into a story and a healing journey.
The original design relied on computer vision to read a photo of the wound and a large language model to write a story around it. That architecture needed 12 to 18 months of machine learning engineering and heavy medical-AI compliance work before the app could launch.
The solution
We kept the experience and replaced the expensive intelligence layer underneath it.
Swap computer vision for visual matching
Instead of analysing a wound photo, the app shows a curated library of illustrated wound cards. The child or parent picks the closest match, which needs no model training and no image processing.
Replace live AI with a story engine
A rule-based engine personalises 350 pre-authored story templates using simple JSON rules. Each story feels tailored without a single live AI call.
Keep the game intact
The full 7-day gamified healing journey stayed in the product end to end, so the redesign cost the experience nothing.
Build on a proven stack
We delivered the app on React Native and Firebase, with Contentful for content, WebAR, Mixpanel analytics and a Node.js backend.
- React Native
- Firebase
- Contentful
- WebAR
- Mixpanel
- JSON rules
- Node.js
The result
Build cost came down sharply because the project no longer needed custom computer vision training or LLM fine-tuning. The 12 to 18 months of machine learning work in the original plan came off the roadmap.
Children still get a personalised story and the full 7-day healing journey. With 350 pre-authored templates and a rules engine doing the work, the app runs with zero live AI calls.
- No custom computer vision model to train or maintain
- No LLM fine-tuning and no live AI calls at runtime
- 350 personalised story templates, managed as content
- The complete 7-day gamification loop preserved
An experience can feel intelligent without live AI behind every screen. Matching the architecture to what the product actually needed took a year or more of machine learning work off the critical path.