I just shipped a production-grade, cross-platform health application in three days: backend, web dashboard, mobile app, AI-powered insights, 274 automated tests, and CI/CD pipelines.
The Project: VitalsBase
The idea was simple: patients get vitals readings (glucose, blood pressure, heart rate) but have zero context for what those numbers mean. VitalsBase closes that gap. It tracks glucose readings against clinically recognized reference ranges, adapts those ranges based on meal context, visualizes trends with color-coded charts, and uses Claude’s AI to explain what the data means in plain language.
The result is a three-platform product: a FastAPI backend with PostgreSQL, a Next.js web dashboard with interactive Recharts visualizations, and a Flutter mobile app for iOS and Android. All sharing a single API and the same AI intelligence layer.
The AI Skills That Made This Possible
This wasn’t about writing code faster. It was about applying a set of AI skills I’ve been deliberately building over the past year. Skills that turned what would have been a months-long team effort into a focused 3-day sprint.
Before a single line of code was written, I used Claude to work through every major decision: database schema, API design, state management patterns, navigation structures, charting approaches. The ability to have a deep architectural conversation with AI, pressure-test decisions, and arrive at a documented plan before touching code is a skill most developers haven’t developed yet. It’s the highest-leverage AI skill I’ve found.
There’s a massive difference between asking AI to “build me a health app” and asking it to “create a Riverpod FutureProvider that calls a Dio API client to fetch glucose stats, returning a typed GlucoseStats model with null-safe JSON parsing.” The specificity of the prompt determines the quality of the output. I’ve learned to decompose large systems into precise, contextual prompts that produce production-quality code on the first pass.
I used two instances of Claude working in parallel: Claude.ai for strategy, architecture, and documentation, and Claude Code in the terminal for implementation. Knowing when to use which AI tool, and how to feed context between them, is a workflow skill that compounds with every project.
The 274 tests across three platforms weren’t afterthoughts. They were generated alongside the code, with AI writing tests that catch real edge cases. AI-assisted testing, accessibility auditing, and performance optimization are what turn “it works on my machine” into “it works everywhere, for everyone.”
Integrating Claude’s API into the product itself, with structured prompting, rate limiting, response caching, mandatory disclaimers, and cost management, required understanding how to build AI into products responsibly. This is the skill gap most organizations are facing right now: not whether to use AI, but how to ship it safely.
What 3 Days Actually Looked Like
Day one was all backend: the complete API, database schema, AI integration, seed data, and test suite. By end of day I had a fully documented, working API ready for anything the frontend needed.
Day two I moved to the frontend. Seven screens for the web dashboard, AI insights verified against real data, and then the entire Flutter mobile app across six screens. All three platforms were running against the same backend by evening.
Day three was about hardening: accessibility audit to WCAG AA, performance work, 274 automated tests across all platforms, onboarding flows, and CI/CD pipelines with GitHub Actions deploying to Railway and Vercel.
The Bigger Picture
After 15+ years in enterprise technology, I’ve seen plenty of hype cycles. This one is different. The “AI writes code” framing is surface level. The deeper change is the economics of building software. One person with the right AI skills can now deliver what used to require a team and a quarter.
AI amplifies expertise. It doesn’t replace it. Every architectural decision behind VitalsBase drew on years of enterprise experience: database design patterns, API conventions, accessibility standards, deployment architecture. The AI made the execution faster. The human made the decisions that determined whether the result would be good.
The professionals who will define the next era aren’t the ones who can prompt AI the best. They’re the ones who know what to build, why to build it, and how to architect it, then use AI to execute faster.
What’s Next
VitalsBase is expanding into multi-vital tracking, blood pressure, heart rate, SpO2, temperature. The architecture was designed for it from day one. I’ll be writing about that build as it happens.
Find me on LinkedIn if you’re building something similar.
This is part of my ongoing series on building with AI. Previous posts cover getting started with Claude Code, building a web app with Claude Code, and vibe coding a WordPress theme.