AI CRYPTO TRADING BOTS / ALGORITHMIC TRADING
PROPELOO engineers algorithmic crypto trading systems — from strategy codification and backtesting infrastructure through live execution engines, risk management, exchange API integration and the monitoring that keeps automated capital deployment under control. Trading bots are not magic. They are strategy implementations with robust execution and explicit risk controls.
The most common failure mode in algorithmic crypto trading is not a bad strategy — it is a strategy with no risk controls deployed in live markets. A trend-following strategy that worked in 2021 bull markets but has no drawdown stop will keep trading in a 2022 bear market. A market-making bot with no position inventory limits will accumulate directional risk during volatile periods. A cross-exchange arbitrage bot with no latency monitoring will execute at prices that have already moved. Trading system engineering requires treating risk management as equal in importance to signal generation. PROPELOO builds trading systems where the risk controls are as sophisticated as the strategy.
WebSocket connections to exchange market data feeds, OHLCV storage (TimescaleDB), order book depth capture, on-chain DEX data via The Graph, news/social sentiment feeds.
Vectorised or event-driven backtesting with realistic transaction cost modelling (slippage, fees, funding rates), walk-forward optimisation, out-of-sample validation and performance metrics.
Exchange API integration (ccxt library), smart order routing, iceberg orders for large positions, TWAP/VWAP execution for size, maker/taker optimisation and fill monitoring.
Per-trade position size limits, portfolio exposure limits, maximum drawdown circuit breakers, correlation-based risk, VaR calculation and automatic position liquidation on breach.
On-chain DEX execution via viem/ethers.js — Uniswap, Curve, Balancer. Flash loan arbitrage, sandwich attack defence, MEV-protected transaction submission (Flashbots).
Real-time P&L tracking, position inventory display, execution quality metrics (slippage vs expected), fill rate monitoring, strategy performance vs benchmark and automated alerts.
Strategy performance in backtests is hypothesis, not evidence. Live performance with proper risk controls is evidence.
A strategy with 20 parameters optimised on historical data will look excellent in backtest and fail in live trading because it has learned noise, not signal. Walk-forward validation (train on rolling window, test on out-of-sample) and parameter sensitivity analysis (how much does performance change if parameters shift slightly?) are required quality gates before live deployment.
Axiom:
A strategy that generates a 20bps edge per trade but costs 15bps in slippage and fees captures only 5bps. Execution quality — order type selection, timing within the bar, exchange venue routing — is where strategies either capture their theoretical edge or give most of it away. Measure actual slippage per trade from day one.
Axiom:
A strategy without a drawdown stop is not a trading system — it is a gamble on the strategy never being wrong. Maximum drawdown limits, daily loss limits, position concentration limits and automatic position reduction on adverse moves are not optional risk controls. They are what differentiates a trading system from a speculation.
Axiom:
Every strategy should run in paper trading mode (simulated execution on live data) before deploying real capital. Paper trading catches: bugs in order management logic, incorrect fee assumptions, edge cases in position sizing, API behaviour differences from what backtesting assumed.
Axiom:
Impact: Strategy selection depends on market regime, capital size and risk tolerance. Arbitrage strategies have the most deterministic edge but require fast execution. Trend/mean reversion require drawdown tolerance.
Impact: CEX for most strategies — better liquidity, lower execution costs, established APIs. DEX for on-chain strategies where trustlessness is required or MEV opportunities justify the on-chain complexity.
Impact: Vectorbt for large-scale backtesting across many assets and parameters (fast vectorised computation). Freqtrade for strategies that benefit from its live trading integration. Custom for complex multi-asset strategies with non-standard execution models.
Impact: ccxt for CEX connectivity — reduces integration complexity significantly. Direct exchange API only when sub-millisecond latency is genuinely required for the strategy.
Impact: Per-trade stops + portfolio-level drawdown limits is the minimum for live trading. Correlation-adjusted risk prevents strategies from having hidden concentration risk.
Impact: VPS or cloud instance (AWS EC2/GCP VM) for most trading systems — sufficient uptime with health monitoring and automatic restart. Co-location only for HFT strategies where microseconds matter.
Multi-exchange momentum system with ATR-based position sizing, pyramiding rules, trailing stops and portfolio-level exposure management.
Pairs or basket trading system — cointegration analysis, mean reversion entries, dynamic hedge ratio calculation and execution optimisation.
Cross-DEX arbitrage with flash loan execution, MEV-protected submission via Flashbots, gas cost optimisation and profitability simulation.
Automated market making on CEX or DEX — bid/ask spread management, inventory control, adverse selection detection and profitability per token.
Vectorised backtesting infrastructure for strategy research — walk-forward optimisation, Sharpe/Sortino analysis, drawdown analysis, transaction cost modelling.
Automated portfolio rebalancing to target weights — tax-aware rebalancing, execution optimisation, drift monitoring and performance attribution.
Stack: ccxt (100+ exchanges), Binance Python SDK, Bybit API, Coinbase Advanced Trade API
Stack: TimescaleDB (time-series), InfluxDB, WebSocket data collectors, The Graph (DEX data)
Stack: Vectorbt, Backtrader, Freqtrade, Custom Python framework, pandas + numpy
Stack: viem, ethers.js, Flashbots (MEV protection), 1inch API (aggregation)
Stack: Python (asyncio), Redis (state), PostgreSQL (trades), AWS EC2, Supervisor/PM2
Stack: Grafana + Prometheus, PagerDuty, Telegram bot alerts, Custom P&L dashboard
Exchange API keys with withdrawal permissions disabled — trading only. IP whitelist on exchange side. Keys in AWS Secrets Manager, never in code or env files.
Trading bots should never have withdrawal permissions. Read + trade permissions only. Even if the system is compromised, the attacker cannot withdraw funds.
Hard-coded maximum loss per day, maximum position size, and maximum order rate. These cannot be overridden by strategy logic. If triggered, system pauses and alerts.
Every order is validated before submission: price within reasonable range of mid-market, size within limits, not duplicate of recent order. Prevents runaway order submission bugs.
Trading system server: no public SSH access (VPN only), firewall rules restricting outbound connections to exchange endpoints only, monitoring for anomalous behaviour.
On-chain trading: private keys in HSM or hardware wallet for signing. ERC-4337 session keys with specific contract/value permissions for automated on-chain execution.
Codify strategy rules precisely — entry/exit conditions, position sizing, risk parameters.
Historical data collection, TimescaleDB setup, live data feeds.
Strategy implementation, backtest with realistic costs, walk-forward validation, parameter sensitivity analysis.
Exchange connectivity, order management, fill handling, execution quality measurement.
Position limits, drawdown stops, circuit breakers, portfolio-level exposure.
Live data, simulated execution — 2-4 weeks to validate live behaviour matches backtest.
Small capital deployment, monitor closely, scale capital as live performance validates strategy.
Walk-forward validation: train strategy on rolling window (e.g., 6 months), test on the next month, advance by one month and repeat. Compare in-sample to out-of-sample Sharpe ratio — significant degradation indicates overfitting. Parameter sensitivity analysis: vary each parameter ±20% and measure performance change. Robust strategies are insensitive to small parameter perturbations.
Slippage is the difference between expected execution price and actual execution price. Causes: market impact (your order moves the price), latency (price moves before your order fills), spread (you always buy the ask and sell the bid). Minimise via: limit orders over market orders, TWAP for large orders, trading in high-liquidity periods, venue selection (use exchange with tightest spreads for your pair).
For personal account trading: generally no license required. For managing third-party capital: requires investment advisor registration in most jurisdictions (RIA in US, FCA registration in UK). For running a trading service (where others subscribe): may require money service business or similar licensing. Regulations vary significantly by jurisdiction and capital size. Legal review before accepting third-party capital is strongly recommended.