qusong0627/QuantMind
QuantMind（量化大脑）开源版是一款面向个人开发者与投研团队的 AI 原生多市场量化交易平台。深度集成微软 Qlib、RD-Agent 因子演化与 TradingAgents 多智能体投研，提供从 300+ 维因子挖掘、13 种机器学习与深度学习模型工场、Optuna 自动调参、Qlib 高性能回测、截面批量推理、7x24 实时舆情情绪分析，到通达信深度联动（板块推送/预警雷达/闪电下单）与实盘模拟交易的完整闭环。全面支持 A股、港股、美股、期货及区块链。系统采用 Docker Compose 一键私有化部署，数据与模型完全本地化，保障策略隐私，功能零门槛无限制。
What it solves
QuantMind is an AI-driven quantitative trading platform designed to lower the barrier to entry for professional quant research. It solves the primary obstacles faced by individual researchers: the difficulty of acquiring and cleaning high-quality financial data, the complexity of building predictive features, and the high technical threshold for training and deploying machine learning models for market prediction.
How it works
The platform creates a complete research loop from data acquisition to live trading. It integrates QuantDB, a professional data hub providing pre-cleaned A-share data and over 315 AI factors. Users can train any of 13 built-in AI models (ranging from tree-based models like LightGBM to deep learning architectures like Transformers and TCN) using a visual interface. The system supports automated hyperparameter tuning via Optuna and can offload heavy training tasks to remote GPU clusters (AutoDL). Once trained, models are automatically registered and used for multi-market inference to generate trading signals, which can then be backtested via Microsoft's Qlib framework or pushed directly to trading software like TongDaXin.
Who it’s for
It is built for individual quantitative researchers, academic teams, and professional institutions looking to validate strategy prototypes or perform secondary development in the field of AI-driven quantitative trading.
Highlights
- Extensive Model Library: 13 professional ML/DL models including LightGBM, XGBoost, CatBoost, GRU, LSTM, Transformer, and TabNet.
- Professional Data Integration: Direct access to QuantDB with 5000+ A-shares, 20+ years of history, and 151+ dimensional feature engineering.
- AI-Powered Research: Includes RD-Agent for automated factor mining, TradingAgents for multi-agent research reports, and QuantBot for natural language interaction.
- Institutional Workflow: Features Walk-Forward Analysis (WFA), Stacking ensemble methods, and production monitoring with real-time Rank IC backfilling.
- Remote GPU Training: One-click deployment of training tasks to AutoDL GPU clusters to accelerate deep learning model convergence.
Related
- Project
- Project
- Project
- Project
- Project