The Brief
Built as a personal tool for one experienced crypto day trader: scan OKX perpetual futures markets and turn a pile of noisy technical data into a compact, three-part decision checklist — crowding, then reversal, then execution — instead of a black-box score claiming to be a win probability.
The Approach
A Flutter app (Android and web) that reads only public OKX market data — no API key required, and it is structurally unable to place a real order. Signal inputs span account ratio, funding rate, open interest, taker buy/sell flow, a live order-book snapshot, and a standard technical-indicator stack (EMA, RSI, MACD, ATR, anchored VWAP).
Where the Real Engineering Is
The interesting part isn't the indicators — it's the background infrastructure. A Cloudflare Cron Trigger rotates a scanner through the full market universe in batches every few minutes, automatically opens simulated "paper" positions when a strategy's rules are met, and tracks them to close using real market candles stored in Cloudflare D1 — entirely independent of whether the app or a browser is open. A read-only Telegram bot reports status on command. A separate research pipeline correlates trading activity against macro data (equities, gold, FX) and scheduled economic events (CPI, NFP, Fed decisions), logging results honestly even when they come back inconclusive rather than cherry-picking the correlation that looks good.
Built to Fail Honestly
Older strategy versions are frozen from new entries rather than deleted, so a bad idea's historical results stay auditable instead of quietly disappearing. The tool is explicit about its own limitations: signal thresholds are heuristic and not yet calibrated from recorded outcomes, and the order-book snapshot can be spoofed. This is a paper-trading research tool, not financial advice, and it does not execute live trades.
The Outcome
A working, live dashboard plus a rigorous, automated pipeline for testing a trading hypothesis without risking capital — background Cron automation, an edge database, real-time data ingestion, and a bot integration that all keep running unattended. That's the same reliability-under-the-hood discipline that carries over directly into SaaS Platforms work that needs automation running correctly whether or not anyone's watching.