Building OnTempo: Shipping a Mobile App With AI (and Experience) as a Force Multiplier
When a bloated BPM app slowed down my class prep, I built my own. OnTempo is a fast, ad-free BPM counter created with AI tools like Cursor, Claude, and ChatGPT. This post breaks down how I shipped it to both app stores and what I learned along the way.
Outside of my work at Summit Foundry Group, my side hustle is coaching group fitness at Life Time. Every week I’m building playlists for several different formats, and I run into the same friction point over and over: I need a fast, accurate BPM counter that lets me find the feel of the song.
For years, I relied on a handful of lightweight apps to get it done. Then one day, while prepping a cycling class, I opened the trusty BPM app I’d used forever… and it was unrecognizable. It launched straight into a paywall modal. The viewport shrank under an ad banner. It took several seconds just to start tapping. The whole thing felt like a haunted house of growth hacking.
My immediate reaction wasn’t even frustration — it was clarity: I can build my own. No ads, no tracking, fast, clean, fun. Honestly, I’d happily pay five dollars for that myself.
A few weeks later, OnTempo shipped. It’s now live on both app stores:
https://ontempo.app
This post is the story of how it happened, what I learned building an app almost entirely with AI tools, and why experience still matters more than ever.
Why OnTempo Exists
OnTempo is intentionally simple. It launches instantly, lets you tap a rhythm, and gives you a BPM measurement that matches the feel of the song. No ads, no trackers, no nudges to upgrade — the base version is complete the moment you install it.
I deliberately left out anything that would distract from the core loop. There’s no microphone beat detection, no metronome, no playlist builder, no cloud sync. Just a tap interaction designed for coaches, athletes, and anyone who cares about timing. There may eventually be a “pro tier,” but v1 needed to feel self-contained.
That constraint also kept the product grounded and gave the AI far less surface area to wander off into complexity for its own sake.
How 80% Was Built in a Single Day
The surprising part of this project is that the majority of OnTempo — the UI, the tap logic, the flow, the polish — came together in roughly a day.
I started with Cursor’s Plan Mode and a simple prompt describing what I wanted: a fast, clean BPM counter with optional precision settings (4, 8, or 16 taps). Cursor generated a solid plan, scaffolded the project, and produced a working version almost immediately. From there, iteration happened conversationally — tightening behaviors, simplifying flows, and eliminating friction.
Claude Code and ChatGPT supported the process by refining logic, improving performance, and helping identify oddities. The experience felt less like “AI made an app for me” and more like pairing with a very fast junior developer who benefits from tight direction.
You still need product taste. You still need a vision of “done.” But once you have those, the acceleration is real.
The Final 20%: All the Unsexy but Critical Work
The remaining 20% took another week, and none of it was glamorous. I added anonymous analytics with PostHog and had to aggressively strip out any identifiable data — IP addresses, locations, and anything else that could fingerprint a device. The SDK collects a lot by default, so making it truly anonymous was more about careful subtraction than addition.
There were also the usual small UX papercuts: startup timing, subtle animation adjustments, and ensuring tap accuracy felt predictable. AI gets close on these, but finishing them requires human judgment.
Where the Real Time Went: Tooling and Automation
Building the core app was fast. Shipping it was not. The real time sink was everything “around” the app:
- CI/CD pipelines
- Signing workflows and provisioning profiles
- Android keystores
- Translations
- Store metadata
- Screenshots for every device size, orientation, and platform
My final workflow uses GitHub Actions with a self-hosted Mac Mini runner at home. It’s fast, cost-effective, and fully under my control.
The screenshot workflow was the biggest challenge. Detox doesn’t work with Expo 54, and native capture required deep linking or adding a special “screenshot mode” to the app. Eventually I landed on Maestro, which uses a clean DSL and hooks into React Native accessibility tags. After that, the process was smooth — even if a single full pass still takes about an hour.
App Store Chaos
The apps were approved on the first submission, which I credit both to years of experience shipping consumer mobile apps and to a pre-submission code review with Claude. But the surrounding bureaucracy was another story.
The DUNS number process was the worst part. My first application was rejected without explanation; the second attempt — submitted through an obscure link inside Apple's developer workflow — was accepted within a day.
To make matters more interesting, I ended up paying for two Apple Developer accounts because support stopped responding halfway through a migration request. These are the kinds of things AI can’t help with — the human systems are still the messy ones.
When Experience Saves You (and the AI)
Even with great models, there were moments where experience mattered.
The AI kept pushing Expo’s cloud builds. The free tier queues can be long, and the paid tier gets expensive quickly. I knew I wanted local builds for speed and cost control, so I ignored the AI and built the workflow I knew would serve me better.
Another time, AI insisted Detox was the right choice for automated screenshots, even while contradicting itself about Expo compatibility. Eventually I reframed the question — “What frameworks can automate mobile app screenshots?” — and Maestro surfaced immediately. A few minutes later, I had a working proof of concept.
These moments reinforced the reality: AI is powerful, but experience is what keeps it on track.
Mentoring the AI: The Real Skill
There’s a misconception that AI can replace developers. It doesn’t. What it does is give a developer leverage — the ability to explore more iterations, remove mechanical overhead, and accelerate the parts of development that don’t require deep intuition.
But AI doesn’t have taste. It doesn’t understand what “good” feels like. It doesn’t think about long-term maintainability unless you explicitly teach it. It doesn’t reason about tradeoffs. And when it hallucinates, inexperienced developers often follow it into dead ends.
Most of my interaction with the models was about direction: nudging plans, overriding complexity, correcting misunderstandings, or supplying the answer outright. Sometimes I even used one model to draft prompts for another.
In that sense, AI is a multiplier of your experience — not a substitute for it.
Why This Was a Great First SFG Use Case
OnTempo isn’t the type of project clients typically bring to Summit Foundry Group. It’s small, simple, and not tied to a business model. But the process mirrored the same patterns SFG uses with founders and investors every day: clarity, speed, iteration, strong technical foundations, and pragmatic use of AI to accelerate without adding noise.
It was a clean testbed for modern development workflows and a reminder that small, personal projects can sharpen the same muscles used on large-scale technical engagements.
V1 Philosophy and What Comes Next
Version 1 is intentionally simple. It’s complete the moment you install it. Over time, I may add features, a pro tier, or deeper tooling for coaches and DJs. Or I might experiment for the fun of it — that’s the beauty of a hobby app built with professional discipline.
Looking Back After Launch
Seeing OnTempo live on both app stores has been gratifying. I’ve shipped major consumer apps before, but this one is different — it’s small, personal, and built with tools that didn’t exist a few years ago.
The feedback has been great. Coaches and friends have reached out with support. People downloaded it, used it, and paid for it. Now the real fun begins: improving it based on how people actually use it.
If you’re curious how AI-accelerated builds can support your roadmap or technical strategy, that’s exactly the space Summit Foundry Group operates in. Sometimes clarity in complexity starts with a simple rhythm: tap, tap, tap, tap.