Case study · Client: Antoinette (own product, built with Jan Friedmann)
Antoinette: Published on the App Store and Google Play
Our multi-agent social assistant as a native iOS and Android app: talk or type to your agent, then swipe through a daily brief of plans, invites and check-in nudges prepared by your friends' agents.
Live on iOS and Android from one codebase; releases are built and submitted to the stores automatically.
TL;DR
Live on the Apple App Store and Google Play, one Expo / React Native codebase, currently v1.0.32.
Built with Jan Friedmann as lead engineer (about 1,380 of the project's 1,567 commits). Konstantin on product, design and testing.
Fully automated releases: GitHub Actions builds, submits to Apple review and pushes Android tracks.
Team
Who
Role
Jan Friedmann
Lead engineer: backend, agents, app, infrastructure and release automation
Konstantin Hilverth
Product, UX, testing, content, go-to-market and EU grant application
What the app does
Capture: hold to talk or type. The Fetcher agent turns it into a plan, an update or a feeling and asks follow-up questions if something's missing.
Daily brief: a swipeable inbox. Accept or decline plans and invites, see what friends are up to, and get check-in suggestions.
Social layer: friends, sharing circles (Close friends / Family / Acquaintances), comments, reactions, photos and mentions.
Events: create events, let friends vote on time options, co-admins, and reposts that widen the invite circle.
Sign-in: Apple, Google or email code. Contacts matching checks who's already on Antoinette and never stores your contact list.
Publishing to iOS and Android
1
One codebase. Expo + React Native + TypeScript + NativeWind (Tailwind for native).
2
Two kinds of update. JavaScript-only changes ship over the air through Expo Updates, with no store review needed. Native changes trigger a full store build. A fingerprint policy decides which applies.
3
Automated release workflow (GitHub Actions):
merge main into prod → deploy the backend to Cloud Run,
build iOS and Android with Expo EAS,
create the App Store version automatically (even on a cold start) and submit for Apple review,
push the Android build to Google Play testing tracks,
broadcast the release to users.
4
A demo account for reviewers. Apple reviewers sign in to a dedicated demo database schema with realistic seeded data, so the app never looks empty in review.
5
Emergency brake. One workflow withdraws a build from App Store review and Play alpha if something's wrong.
6
Store compliance. Privacy-safe permission texts (photos only when you pick one, contacts never stored), account deletion, privacy policy and terms on the website.
FastAPI on Google Cloud Run (one warm instance, SSE streams for the live brief)
Data
PostgreSQL on Supabase. Separate schemas for prod, demo, test, sandbox and every feature branch
Agents
Fetcher, Observer, Empathy Proxy, Event Curator, Identity and Brief Digest, on a scheduler
Auth
JWT with rotating refresh tokens, email OTP, Sign in with Apple and Google, rate limiting
Ops
Admin dashboard (network view, user stats, prompt sandbox, agent test scenarios), Renovate, cold-start and 401 alerting, one-command rollbacks
Results
Metric
Value
Platforms
iOS + Android, one codebase
Current version
1.0.32
Commits (Mar–Oct 2026)
1,567
First commit → Telegram MVP
under a week
Telegram → App Store submission
~3.5 months (release workflow live June 2026)
Beyond the app
Antoinette is also the basis for an EU funding application under the call PPPA-2026-YOUTH-SOCIAL-MEDIA-DESIGN, which looks at social media designed around real-life connection instead of endless feeds.
