Full stack, from the Flutter app in your hand to the Go services, distributed databases and AI pipelines behind it. Promoted three times in five years. Your next engineer is scrolling past this sentence.
Every number here comes from production systems I designed, built or led.
Most engineers pick a layer. I've shipped every layer of the same products to production: the Flutter app, the Go services behind it, the distributed data underneath, and the AI that feeds it content.
Built Wordy from zero in Flutter and launched it on Android and iOS. Shipped Leagues, MiniGame and Daily Goals to 8M+ users across Flutter, Java and Objective-C, and built Daily Themed Crossword for Facebook Instant Games with Meta's team.
REST APIs and microservices for a puzzle portfolio played by millions, plus event-driven flows on Kafka. Backend and Cloudflare CDN tuning delivered +15% engagement, +10% retention and −12% load time.
A leaderboard platform on Redis + ScyllaDB adopted by 10 games (~3M daily players), and a coordination-free ID service for 1M+ DAU. Partition-aware models and caching that absorb thousands of writes per second.
An async orchestration platform across OpenAI and Gemini (5× images, 2.4× puzzles), and a private RAG assistant on Qdrant + Qwen3 (~30% support efficiency).
Real-time ranking at scale is a system I've already built: the Go + Redis + ScyllaDB platform behind 10 games. Here it is, applied to me. Hit ▲ to vote for what matters to your team.
Simplified, interactive versions of real systems I've shipped to production. Click everything.
One shared platform for global, regional and event rankings, replacing a separate implementation in every game. Clients batch score syncs every 30 seconds, Redis sorted sets serve ranks hot, and ScyllaDB keeps it all durable on partition-aware models. I built Leagues on top of it in Flutter.
↳ Tap Solve a puzzle on the phone and watch your score travel through the stack. Can you reach #1?
Every server owns a non-overlapping range and mints IDs locally. When a server dies, its leftover range is thrown away and the restart claims a fresh one, so an ID can never repeat. This model issues IDs to users across every game, device and login provider.
I saw that content production was the real bottleneck, so I built an async orchestration platform that generates, validates and publishes puzzles and images across several AI providers. The content team now ships to production on its own.
BGE-M3 embeddings, Qdrant vector search, cross-encoder reranking and a self-hosted Qwen3 8B on Ollama. It drafts replies from past tickets, and every byte stays on company infrastructure.
Replaced per-game leaderboards with one shared platform adopted by 10 games.
Saw that content, not code, was slowing launches, and built the platform that fixed it.
Led the backend platform for Crossword Go, played by 400K people every day.
Built Wordy from scratch in Flutter and launched it on Android and iOS.
Leading the engineering teams behind Crossword Go (400K daily players) and Pawzzle!
Leagues, MiniGame and Daily Goals across Go, Flutter, Android and iOS for 8M+ users.
These aren't side projects. They're live games, played by hundreds of thousands of people every day, running on backends I built and led.
Team lead. I lead the engineering team and own the backend platform for content, gameplay and real-time player experiences, including leaderboards and events.
Shipped Leagues, MiniGame and Daily Goals across Go, Flutter, Android and iOS, and built the Facebook Instant Games version with Meta.
A daily word puzzle I built from scratch in Flutter and launched on Android and iOS, end to end from client to backend.
Word puzzles on the shared backend platform for content, gameplay and real-time player features that I led.
I lead the engineering team behind this logic puzzle game, owning technical direction and delivery across client and backend.
Flutter, Android and iOS on the front; Go, Redis and ScyllaDB on the back; AI pipelines feeding it all. A feature goes from screen to database without a hand-off.
Partition-aware models, caching that holds up under thousands of writes per second, and failure modes I design for up front, not after the outage.
Not demos. Production pipelines and private RAG with measurable gains: 5×, 2.4×, ~30%.
I find the bottleneck nobody filed a ticket for, then take it from architecture to production and across product, content and art teams.
Every system I design ends in a consistent state. So does this page. Solve it, or skip straight to the buttons. I won't tell.