Ceryo

The context

Ceryo is a super-app: a modular mobile application bringing together planning, learning, a content library, real-time messaging, budget, health, travel and places, a rich editor and an AI assistant, all in an Apple-grade, fluid, offline-first experience where every screen must load instantly and behave like a system app.

My role

Solo, end-to-end product design, architecture and full-stack development: Rails backend (API, business services, jobs, real time, AI and payment integrations), React Native mobile app (screens, navigation, data hooks, native health and in-app-purchase integrations), web platform and showcase, design system, tests, CI and self-hosted containerized deployment.

Key features

• Augmented search engine: real-time web search, multi-source synthesis and a personal encyclopedia with semantic search and an automated editorial pipeline.
• Ceryo Assistant: an AI copilot with five LLMs to choose from, in orchestrator mode (acting across every module) and architect mode (project planning).
• Arena, gamified learning: quizzes, real-time duels, scheduled battle royales, teams, ELO, leagues, XP, streaks, badges and an evolving mascot.
• Spaces, themed communities: posting, feed, reactions, polls, roles and moderation, monetization and live events.
• Budget: expense tracking with AI categorization, recurring subscriptions, wallets and savings goals.
• Sport & Health: adaptive programs, body-composition tracking and wearable sync (Apple HealthKit, Android Health Connect).
• Travel & Studio: day-by-day itineraries, travel budget, and a unified creation workshop (notes, documents, spreadsheets, canvas) with an embedded rich editor.
• Cross-cutting: real-time chat and messaging, missions and trophies, push notifications, payments (Stripe subscriptions and in-app purchases), roles and permissions, FR→EN i18n.

Technical challenges

• Unified multi-provider LLM gateway: a single interface for five providers with different formats, normalizing messages, function-calling and SSE streaming, routing by task type, an automatic fallback chain and per-request cost estimation.
• Multi-domain real time: nine Action Cable channels covering presence, messaging, community events and synchronized competitive games, backed by Redis.
• Mobile 'Batch Snapshot' pattern: each screen consumes a single endpoint returning all of its data, with a dedicated hook and skeleton, on-touch prefetch and a per-screen freshness policy: zero waterfall loading.
• In-database semantic search: pgvector enabled and exploited (1536-dimension embeddings, cosine IVFFlat index) for the encyclopedia and search, with no external vector engine.
• Autonomous editorial automation: around sixty recurring jobs orchestrate ingestion, synthesis, sport sync and event lifecycles, designed to run unattended.
• Reliability of third-party data pipelines: robustness of multi-provider aggregation (cross-provider deduplication, alias resolution, live scores, handling of free API tiers).

Scale

A platform-sized product, built solo.

• A rich mobile app backed by a full server: platform-scale ambition, carried solo.
• Brings together in a single app worlds that usually live apart: budget and wealth, communities, sport and coaching.
• Around twenty business domains treated in depth, not a thin shell.
• Available on TestFlight and exercised by real users, on infrastructure owned end to end.

Results

A complete mobile app, in early access on TestFlight and ready for the App Store.

• Apple-grade mobile experience: smooth animations, haptic feedback and a canonical art direction carried across every screen.
• End-to-end full-stack command: React Native, real-time Rails and a sophisticated AI layer, all built solo.