← All case studies Case study · iOS

Signal10: from web app to native iPhone app

A working learning-discovery product became a tested, native iPhone app built on its existing intelligence, not a second copy of it.

Platform iPhone, iOS 18+ Stack SwiftUI on an existing Node.js service Method RepoGenics

Signal10 results screen showing a Start here card and the learning path

1 day

From the brief to a tested app running on a real iPhone

Written success criteria, each one tested

Lines of server code changed to support the app

API keys or secrets inside the iPhone app

The request

“Make it feel built for iPhone, without reinventing it.”

YouTube has the knowledge. Signal10 gives you the path.

Signal10 answers one question: what are the best videos to watch to learn this subject? It searches YouTube, cuts the noise and duplicates, ranks on more than view counts, and turns the best ten into a learning path. It already worked as a web app.

The brief was to make it a production-quality native iPhone app, with firm boundaries:

Treat the existing product as the source of truth: same ranking, reasons, learning path and backend.
Never ship an API key or model secret inside the app.
Keep the AI provider replaceable, so the app never depends on one model vendor.
Keep it simple: one question, “What do you want to learn?”, and an obvious next video.

How it was delivered

The Repogenic Loop, applied to one app

Every step left evidence in the repository, so the speed came with proof.

Intent

The brief became a one-page intent with 10 numbered, testable success criteria, written after reading the whole existing repository first.

Variation

AI-assisted engineering produced the SwiftUI client, its models and its tests, reusing the existing service’s API instead of rebuilding its logic.

Selection

Failing tests were committed first, then code until they passed, then real searches and an automated walkthrough on a simulated iPhone.

Inheritance

Each success merged with its evidence: screenshots, a build-and-run guide, and notes so the next feature starts from what works.

Guardrails

Secret scanning, a no-keys-in-the-app test, untrusted YouTube text rendered as plain text, and the AI boundary kept on the server.

The output

A learning path you can work through

Type a topic, start watching. The app remembers where you are, so coming back always shows what’s next.

Home screen asking What do you want to learn? — **One question**No account and no YouTube sign-in. Recent searches and examples one tap away.

Results with Start here and the learning path — **A curated Top 10**Start here, then stages from foundation to practical, with a switch to the plain ranking.

Video playing with Why Signal10 picked it — **Watch and understand**Plays in YouTube's own player, with the reason and five ranking signals behind each pick.

Up next moved to step 2 with progress shown — **Pick up where you left off**Mark watched, play next. Progress and Up next are remembered on the phone.

Home screen with Favorites — **Favorites**Star the videos worth keeping and replay them from Home.

Settings screen — **Settings**Start view, autoplay, clearing your data, account and about.

Help screen — **Help and how-to**Plain-language steps, and how videos are actually picked.

Signal10

App icon and wordmark carried over from the existing brand.

The evidence

Proven, not just demoed

Tests written before the code	Each feature began with a commit holding only failing tests and the written intent. The git history shows the tests came first.
30 unit tests, 2 end-to-end tests	Decoding real server replies, sign-in, error handling, progress, favorites, settings and player safety. End-to-end tests sign in, search, play, mark watched, continue the path, save a favorite, and open Settings and Help, with a screenshot at each step.
Existing product unchanged	All 79 existing service tests still pass. The web app and its live deployment were left untouched.
No secrets in the app	The built app was scanned for keys and credentials, with no matches. A test fails the build if a key ever appears in the app’s settings.
Running on real hardware	Signed, installed and used on an iPhone 17. Feedback from the device turned into two quick, tested layout improvements.
Second feature, same discipline	Settings, Help and Favorites went through the same loop: an 8-criterion intent, failing tests, code, live end-to-end proof.

Why it matters

Speed is only valuable with proof

Reuse, not rewrite

The iPhone app consumes the product’s existing intelligence. One source of truth for ranking and AI behaviour means one place to improve it, and nothing drifts between web and phone.

Safe by design

Secrets stay on the server. The AI provider sits behind the service boundary, so it can move between model vendors, or route by cost, without an app update.

Evidence a buyer can check

Intent, failing tests, passing tests, screenshots and device builds are all in the repository. That is the RepoGenics difference: faster delivery that holds up to scrutiny.

SwiftUISwift 6Swift TestingXCUITestXcodeNode.js serviceAWS LambdaYouTube embedded playerIntent-driven engineeringAI-assisted TDD

Have an idea for an iPhone app?

From intent to a tested app on your own phone, then on to App Store submission in about five days, with the evidence to show for every step.

Talk to Repogenic