I have 21 days to ship a production-ready crypto alpha engine and need a senior Python/ML quant developer who can move fast and communicate clearly (4 h overlap with EST). The MVP is organised into five modules, but per my roadmap the ML piece comes first; everything else hangs off a clean, walk-forward-safe model. DATA Live market data must stream in via CCXT Pro websockets from Binance, Bybit, OKX and dYdX, yet Hyperliquid depth and trades are the non-negotiable core. Capture 1 s OHLCV plus full L2, cache intraday in Redis and persist at least six months to Parquet. FEATURES Enrich every bar with 50 + TA-Lib studies. Moving Average, RSI and Volume are the key signals I monitor, alongside on-chain metrics and Grok sentiment. Persist engineered features to an HDF5 store for rapid sampling. ML (CRITICAL FIRST) • Daily LightGBM retrain must finish in under two minutes on a mid-tier GPU/CPU • Track experiments with MLflow and emit SHAP explanations to JSON • Absolutely no look-ahead bias—be ready to explain your defence strategy in the bid • Code should default to asyncio and type-hinted style EXECUTION Convert model signals to paper or live orders with a 1 % max risk per trade, slippage guard and Telegram push alerts. DASHBOARD A Streamlit board updates P&L and the equity curve in real time. STACK & CI Python 3.11, Docker, asyncio, pytest with ≥ 95 % coverage, GitHub Actions for CI. MILESTONES M1 Data pipeline, feature store, unit tests M2 ML module and back-test of my three strategy specs M3 Execution layer, Streamlit dashboard, inline docs BID CHECKLIST • One-page PDF back-test showing Sharpe > 1.8 • Public Git repo with Dockerfile that reproduces results • Written answer to “How to avoid lookahead bias?” Daily two-minute Loom updates, Slack comms, NDA and full IP transfer are required. Generic copy-paste proposals will be declined immediately.