Oceano di Versi

What it is

Oceano di Versi is an iOS social platform built exclusively for poetry. It starts from the idea that verses deserve a space of their own, away from the noise of general-purpose social networks: a moderated environment that encourages daily creative expression and rewards consistency.

The app is fully localized in six languages — Italian, English, German, Spanish, French and Brazilian Portuguese — and published on the App Store.

Core features

• Dual feed — a global feed to discover new poets and a following feed for favorite authors.
• Emotion tags — every poem is tagged with an emotion (Love, Melancholy, Joy, Anger, Hope, Loneliness, Nostalgia, Peace, Passion, Pain), so readers can browse by mood.
• Rich editor — compose with optional cover images, save drafts, publish publicly or keep private.
• Social interactions — like, comment, follow and share poems via deep links with Open Graph metadata for rich previews outside the app.
• Full-text search — users and poems indexed with Algolia for instant author and verse discovery.
• Daily prompt — every day at midnight (Europe/Rome) Gemini generates an introspective question in all supported languages, pinned at the top of the feed.
• Monthly leaderboard — +1 point per day with at least one public poem, −3 for every missed day. The monthly winner is featured at the top of the public feed for the first 7 days of the following month.
• AI moderation — every poem, comment and profile photo is reviewed by Gemini 2.5 Flash before publication, with an admin panel for human review of flagged content.
• Reports & flagging — built-in user and content reporting, with automatic notification to moderators.
• Flexible authentication — sign in with Email, Google or Apple.
• Oceano Pro — monthly or yearly subscription via StoreKit 2, with unlimited drafts, a verified badge and premium features.
• Notification center — in-app hub for likes, comments, new followers and moderation updates, with multilingual FCM push notifications.

Architecture

The app is written entirely in SwiftUI (iOS 18.6+) and follows MVVM with Firestore as a single source of truth. All listeners are real-time — no polling.

The backend is made of 22 Cloud Functions Gen 2 in Node.js 22, deployed to europe-west1, handling:

• AI moderation of poems, comments and profile photos;
• push notifications for likes, comments, follows and new posts;
• leaderboard scoring and automatic monthly close;
• multilingual daily prompt generation;
• Algolia sync for full-text search;
• propagation of profile changes (name, photo, premium) across related content;
• Open Graph metadata generation for deep-linked sharing;
• GDPR-compliant account deletion.

Two mirrored environments — production and staging — allow isolated testing on real devices thanks to separate Bundle IDs.

Code examples

Paginated Firestore feed (SwiftUI)

Reactive ViewModel, Firestore query ordered by timestamp with a cursor for pagination:

@Published var globalPoesie: [Poesia] = []
private var lastDoc: DocumentSnapshot?

var query = db.collection("poems")
    .order(by: "timestamp", descending: true)
    .limit(to: 10)

if let cursor = lastDoc {
    query = query.start(afterDocument: cursor)
}

AI moderation with Gemini (Cloud Function)

Poems go through pendingPoems before reaching poems: a Firestore trigger calls Gemini and decides the verdict.

exports.moderatePendingPoem = onDocumentCreated(
  { document: "pendingPoems/{poemId}", secrets: [GEMINI_API_KEY] },
  async (event) => {
    const verdict = await runGeminiModeration(event.data.data());

    if (verdict === "block") return quarantinePoem(event);
    return promotePoemFromPending(event);
  }
);

Full details (fail-open, override, profile photos) in the moderation pipeline doc at the bottom of the page.

Stack

Layer	Technology
iOS app	SwiftUI, Swift 5, iOS 18.6+
Authentication	Firebase Auth (Email, Google, Apple)
Database	Cloud Firestore
Storage	Firebase Storage
Backend	Cloud Functions Gen 2 (Node.js 22)
Search	Algolia
AI moderation	Gemini 2.5 Flash
Push	Firebase Cloud Messaging
Subscriptions	StoreKit 2

Built with AI agents

The entire codebase — iOS app, Cloud Functions, Firestore rules, localizations and this documentation — was written in collaboration with multiple AI agents (Claude and Codex). The human role: defining requirements and constraints, reviewing diffs and architecture, calling non-obvious tradeoffs (e.g. fail-open on moderation, floor-at-0 on scores, stable pointer for the daily prompt). Agents produced implementations, tests, refactors and docs; every change went through review before merge.