Live money / execution discipline
Mercury: private trading engine with AI-assisted research loops
Created a profitable, production-ready private trading engine in a hard, execution-sensitive domain with an objective success metric: PnL. Built and operated a multi-strategy execution engine where capital is distributed across distinct intents rather than a monolithic signal.
Designed deterministic execution and risk controls for live trading while keeping LLM-assisted research and evaluation layers separate from the critical path. The backend lead story: TypeScript, Node.js, NestJS, PostgreSQL/TimescaleDB, Redis, telemetry, deployments, monitoring, incident response, and production ownership.
What it forged: a stricter separation between intelligence, leverage, and control. Research can be exploratory; execution paths need explicit invariants, observability, and the humility to stop improving when the system is already right.
Capital, strategic partnerships, or serious trading-infrastructure conversations: [email protected].
Sovereign systems / founder instinct
Tipsomat -> The Bay: Telegram-native value transfer and governance
Tipsomat started as a minimal Telegram-native token layer: transfer value inside the messenger without fees, then withdraw through off-chain signatures. It was early, rough, and real enough to attract serious investor attention before Telegram’s own crypto direction became obvious.
The Bay extended that instinct into a prototype for a parallel state on top of Telegram: user-created tokens, domain/subdomain bots, token-weighted proposals, AMM exchange, and minimal governance primitives using Telegram as the transport layer.
What it forged: product taste for small missing primitives. Before trying to build a movement, define the minimum technical substrate people can coordinate through: speech, ownership, liquidity, and voting.
Production pressure / external proof
Flagship: 10k+ concurrent users and per-minute portfolio recalculation
Architected the backend for a gamified airdrop product where users managed simulated long/short portfolios on dynamic tokens. Built the state engine recalculating active portfolios every minute against live market data, plus a dual-state anti-cheat model to reduce coordinated exploitation.
Delivered the core API in two months under tight business constraints, supporting weekly feature delivery with API contracts and targeted test coverage. The shipped system helped the company earn another chance through $2.5M in bridge funding.
What it forged: survival-speed backend execution without losing the architecture. Tight runway is not an excuse to skip contracts, state boundaries, or production discipline.
Behavior loops / self-nudging
AnyTracker: behavioral feedback loops in a tiny UI
Built a quantified-self Telegram bot around the contract "just track it": dense in-chat visualization, tokenized activity pricing, and micro-incentives that helped behavior become visible without moralizing.
It started as a personal self-nudging experiment and became a working proof that lightweight feedback loops can create agency where willpower is too blunt. The same idea later connects to nudging, habit formation, and product systems that change behavior through environment design.
What it forged: respect for tiny loops. If the feedback is close enough, honest enough, and cheap enough to maintain, behavior can become safer and more legible.