AI Quantitative Trading Platform Practice: Architecture and Implementation of a Risk-Priority-Based Intelligent Trading System

This article analyzes the GitHub open-source project ai-trading-assistant maintained by dondie52. Built on the risk-priority philosophy, this project constructs an intelligent quantitative trading system with a modular monolithic architecture. It supports multi-timeframe technical analysis, backtesting verification, seamless switching between simulated and live trading, and features enterprise-level security and compliance. Project link: https://github.com/dondie52/ai-trading-assistant, released in June 2026.

Most quantitative systems focus on maximizing returns, but this project takes "survive first, then develop" as its core, placing risk management at the forefront of design. The risk-priority philosophy runs through all levels of the system: multi-factor authentication for user verification, risk rule checks for each transaction, portfolio position management and drawdown control, pursuing optimal risk-adjusted returns within a strict risk budget.

The project adopts a modular monolithic architecture, balancing deployment simplicity and code separation:

• Frontend application (Next.js+React+TypeScript+TailwindCSS): Includes modules such as overview, market, strategy, risk alert, etc., with responsive design adapting to narrow screens.
• API service (NestJS+TypeScript+Prisma ORM): Core modules include authentication (JWT double token + MFA), trading, risk rule engine, audit logs.
• AI signal service (Python+FastAPI): Responsible for machine learning model inference, signal generation, feature calculation.
• Shared packages: Provide technical indicators (SMA/EMA/RSI, etc.), signal scoring, core logic of the risk engine.

The system supports multi-timeframes (1m/5m/15m/1h/4h/1d) to adapt to different trading strategies; real-time market data is pushed via Socket.IO (every 5 seconds), with automatic fallback to HTTP polling when WebSocket is unavailable; each price update triggers calculation of indicators such as trend (SMA/EMA), momentum (RSI/MACD), volatility (Bollinger Bands/ATR), and volume (OBV).

The system provides a complete backtesting framework, supporting historical data backtesting (considering transaction fees and slippage); uses walk-forward analysis to avoid overfitting (select parameters in the training window, evaluate out-of-sample performance in the test window); can export performance reports in CSV/PDF formats, including key indicators such as win rate, profit-loss ratio, Sharpe ratio, Sortino ratio, and maximum drawdown.

Multi-layer risk control covers accounts (daily loss limit, risk budget), strategies (position upper limit, correlation monitoring), and individual transactions (stop-loss/take-profit, position calculation); the Alpaca simulated trading environment is used by default, and live trading is enabled only when ALLOW_ALPACA_LIVE_TRADING=true is configured.

Security aspects: JWT double token authentication, TOTP-based MFA, sensitive data encryption (broker credentials, etc.), complete audit logs (login/strategy modification/order operation, etc.); deployment uses Dockerization (including PostgreSQL/Redis/API/Web/AI service containers), supports environment variable configuration, and provides health checks and monitoring alerts.

Applicable users: Quantitative trading beginners (learning in simulated environment), strategy developers (prototype verification), small trading teams (trading management), educational institutions (teaching platform); usage recommendations: start with simulated trading, strictly implement risk rules, continuously monitor strategy performance, and use audit logs to troubleshoot issues.