Selected work

Things we've built.

Three projects we're proud of, described in enough detail to be real. Client-specific mechanics stay confidential; the architecture doesn't.

What we're building now

01iGaming · Game client

Next-generation slot-game client

We work at the core of a next-generation game client that slot titles are built on for the iGaming industry — three years and counting. It's a 60+ package Nx / TypeScript monorepo on Phaser, where the whole spin lifecycle is modeled as a three-level XState statechart hierarchy: deterministic, free of race conditions, and recoverable mid-session, which is what a regulated game has to guarantee. Animation runs through a fluent GSAP timeline layer that unifies tweens, Spine skeletal animation and component state in one declarative API. We also built the in-browser debug tooling — live statechart inspection, GSAP timeline scrubbing, prefab-pool memory profiling — that the whole team uses to ship new titles.

PhaserXStateGSAPSpineNxRspackTypeScript

02iGaming · AI tooling

AI copilot for game mathematicians

A TypeScript CLI with a React/Ink terminal UI that wraps the whole slot-math workflow: scaffold a model, run Monte Carlo simulations of up to 100M spins across worker threads, and analyze RTP, volatility and hit-frequency. The agent layer is what we're proudest of — chained Claude (Opus / Sonnet) pipelines and a library of 60+ reusable skills that capture how our existing games are built and feed that back to steer new math. It turns a multi-day manual certification review into a repeatable, resumable, cost-tracked pipeline.

TypeScriptClaudeNode.jsInkMonte Carlo

03R&D · Trading intelligence

M.A.N.T.I.S. — news-driven trading signals

An in-house R&D platform that turns financial news into trading signals. It ingests 30+ institutional, crypto and macro sources, then runs an eight-stage pipeline: normalize → deduplicate (Redis + pgvector embeddings) → classify and extract entities with Claude → ensemble sentiment → causal-chain reasoning that models two to three levels of probable consequences per event. Related events cluster into consensus-scored signals, weighted across eight source-quality dimensions, and a closed feedback loop scores every signal against realized 1h / 4h / 24h price moves using immutable anchors — so the system measures whether it was actually right, not whether it merely sounded confident.

TypeScriptClaudePostgreSQLpgvectorBullMQFastifyRedis

More work

iGaming · Frontend

Promo management CRM

A React control center for live casino promotions — tournaments, leaderboards, short races, free-spin and cash-drop campaigns. Operators configure, schedule and monitor mechanics in real time across thousands of concurrent players.

iGaming · Analytics

Real-time analytics dashboard

Streaming analytics for commercial and technical teams — player activity, campaign performance and live KPIs over high-volume event data, built for fast reads and quick strategic decisions.

iGaming · B2B

B2B partner portal

A self-serve B2B portal where partners track client performance, revenue and game metrics — the same high-load React foundation, scoped per partner with role-based access.

Fintech · Banking

Online banking for entrepreneurs

A multilingual online-banking frontend for business clients at a major bank — accounts, payments and statements, with reliability and accessibility treated as first-class requirements.

E-commerce

High-traffic online shops

Server-side-rendered React storefronts on large e-commerce platforms, backed by queue-driven PHP services and Elasticsearch search — tuned for SEO, page speed and conversion under heavy traffic.

Mobile · React Native

Cross-platform mobile apps

React Native apps for iOS and Android — a driver logistics app and a video Q&A platform — with camera, video capture and GraphQL data on a shared cross-platform codebase.

Have a platform or a game to build?

Tell us what you're working on — we reply fast.

Start a project →