What it does Qlib is Microsoft’s open-source quantitative investment platform. It bundles data processing, model training, back-testing, and portfolio optimization into one Python stack. The project covers the whole chain: alpha seeking, risk modeling, portfolio construction, and order execution. It also ships with a “Quant Model Zoo” of published research implementations you can run out of the box.

The interesting bit The platform doesn’t just do supervised learning on price data. It explicitly tackles market dynamics with adaptive concept-drift techniques and reinforcement learning for continuous trading decisions. Recently it added RD-Agent, an LLM-driven system that automates factor mining and model optimization—essentially trying to let an agent do the tedious R&D loop.

Key highlights

• Supports offline research mode and online production serving with automatic model rolling
• Includes implementations of specific published models: HIST, IGMTF, KRNN, Sandwich, TCTS, DoubleEnsemble, ADARNN, and others
• Provides a Point-in-Time database and high-frequency (1-minute) data processing examples
• Modular, loosely-coupled design so components can be used standalone
• Cross-platform: Linux, Windows, macOS

Caveats

• The README is enthusiastic but vague on actual performance numbers or live trading results
• Many features are marked as “coming soon” or under review; the roadmap table is currently empty
• RD-Agent is a separate repo—it’s unclear how tightly integrated it is versus being a companion project

Verdict Worth a look if you’re a quant researcher who wants to test published models without rebuilding the data plumbing. Less useful if you need battle-tested execution infrastructure or clear performance benchmarks out of the box.