A 44,000-line trading workstation that put nine figures of notional volume through three perpetual markets — built end to end.A 44,000-line trading workstation that put nine figures of notional volume through three perpetual markets — built end to end.

The platform's users were acting on automated trade signals but executing by hand across separate tools: one tab for the chart, one for the exchange, a chat window for the signal, a spreadsheet for risk. Every hop cost seconds, and seconds cost money.

The brief was one screen that held market context, account state, analytical signals, order construction, and automation together — and stayed truthful when a data connection dropped mid-session.

Version one turned an existing Zapier-connected signal source into a coherent product experience. Versions two and three moved the core into custom services: normalized exchange data, deterministic analysis, liquidation and cascade logic, and real-time distribution. The fourth version shown here is the terminal that brings all of it to the trader.

I owned the terminal end to end — product requirements, interaction design, the panel system, exchange and wallet integration, execution services, bot controls, guarded API routes, and Cloudflare deployment. Three other developers worked on other parts of the wider platform.

The workspace is a resizable panel grid: chart with indicator controls and interval switching, order book and trade tape, order ticket with cross/isolated margin, multiplier up to 40x, take-profit and stop-loss, and a tabbed account area for positions, balances, open orders, history, and the live signal feed.

Every asynchronous action — order placed, cancelled, bot toggled — confirms through the same toast system, and stale or disconnected feeds say so instead of showing yesterday's price.

Ten Zustand stores coordinate positions, orders, funding, spot prices, trackers, UI state, transfers, and automation across every panel. Hyperliquid connectivity runs through a dual-node client (private node first, public fallback) with WebSocket spot prices, SSE signals, and polled account state.

Wallet authentication is Privy; application data and bot tokens live in Supabase, with 15 edge functions covering token issue and revoke, account reads, signal fetches, command publish and result polling, and a points program.

A separate bot API (13 Next.js routes) lets external agents place, cancel, and close positions through the same guarded services the UI uses. Thirty-three service modules separate exchange access, execution, risk, market data, transfers, bots, and fallback behavior. The whole app deploys to Cloudflare Workers through OpenNext.

Over its lifecycle the platform integrated three perpetuals venues. For one of them I wrote the exchange SDK from scratch — a browser-side signer compiled to WebAssembly plus a typed client — because no usable one existed.

The signal feed in the terminal is the output of the same deterministic analysis pipelines that power the ACP seller agent — one data platform serving a human workstation and autonomous agents alike.

The same discipline carries into the local work: real-time state, honest failure states, and interfaces built for people who don't have time to fight their software.