MovyoHealth · R&D Initiative
Automating competitive research and rapid prototyping using AI tooling
MovyoHealth was an internal R&D initiative focused on GLP-1 medication companion apps. My contribution was not the product strategy itself, but the process used to get there: a research prompt system that turned a team's benchmark work into a structured dataset, and an AI-assisted prototype workflow that compressed delivery from weeks into two working days.
32
Competitive products benchmarked
4
Designers aligned under one research structure
60–70
Interfaces delivered in the prototype
2 days
From synthesis to live prototype
Initiative Context
The challenge was making four designers produce compatible research output across 32 products.
Each designer on the team was expected to research a product space and deliver a working prototype for internal review. The category assigned here was GLP-1 medication companion apps — serving adults on weekly injectable weight-loss medications like Wegovy, Ozempic, Zepbound, and Mounjaro.
The benchmark covered 32 products across GLP-1 tracking tools, nutrition apps, telehealth platforms, and medication management tools. I coordinated the overall research structure while personally doing deep-dive research on 8–10 products.
The core challenge was not simply researching the space. It was making the research output consistent enough to be useful — so it could be synthesized across a four-person team instead of being siloed in individual notes.
Research System
A structured prompt system turned benchmark notes into a composable dataset.
To make four designers produce compatible output across 32 products, I designed a prompt with a fixed schema for every app review: module name, section name, interface name, screenshot reference, and feature description. That turned the benchmark into a structured dataset rather than a set of inconsistent notes.
Why structure mattered
Every reviewed app — from Shotsy and Vivy to MyFitnessPal and Ro — was documented at the same level of granularity in the same format. That made modules, interfaces, and features directly comparable across the benchmark, which made synthesis faster and more reliable.
Lowering the skill floor
The prompt also reduced the skill floor required to contribute useful research. Junior designers could follow the same structure and produce output that was immediately usable — without making their own calls about detail level, naming, or formatting.
Prototype Workflow
Schema first, interfaces second. The architecture became the prompt.
The prototype was delivered through a two-stage AI-assisted process.
Stage 1 — Schema generation
Gemini was used to generate a structured schema from the research output and the product brief: core data objects, their fields, and their relationships. That created a validated product skeleton before any interface generation started.
Stage 2 — Interface generation
The architecture became the basis for a detailed prompt targeting base44. The prompt specified screens, modals, component states, design system values, tone, seed data, and edge cases like empty and loading states. Because those requirements were explicit, the generated prototype required relatively little correction.
The result was a working prototype with ~60–70 interfaces spanning five main screens, onboarding, logging modals, and a progress view — in two working days. The value came from designing the process around AI tooling, not just using AI as a shortcut.
Outcomes
What this demonstrated.
Process design
The research prompt acted as a system that made multiple contributors produce composable output without coordination overhead
Speed with structure
The prototype compressed weeks of interface work into days without dropping component states, seed data, or system coverage
AI as workflow infrastructure
The leverage came from designing the process around AI tooling — not just using AI as a shortcut at the end