Part 4: Multi-Agent Systems
Goal for this stage: Build collaborative architecture. Understand why multiple expert Agents are needed, how to design their division of labor and collaboration, and how Risk Agent has veto power.
Lesson List
| Lesson | Topic | Deliverable |
|---|---|---|
| Lesson 11 | Why Multi-Agent | Multi-Agent Architecture Design |
| Lesson 12 | Regime Detection | Regime Agent |
| Lesson 13 | Regime Misjudgment and Systemic Collapse | Diagnostic Checklist, Degradation Strategy |
| Lesson 14 | LLM Applications in Quant | LLM Enhancement Layer Design |
| Lesson 15 | Risk Control and Money Management | Risk Agent |
| Lesson 16 | Portfolio Construction and Exposure Management | Portfolio Agent, Factor Monitoring |
| Lesson 17 | Online Learning and Strategy Evolution | Evolution Agent |
Background Knowledge
| Document | Description | Suggested Reading Time |
|---|---|---|
| Multi-Agent Framework Comparison | Shannon vs AutoGen vs CrewAI | 15 minutes |
| Quant Open Source Framework Comparison | VectorBT vs Backtrader vs FinRL | 10 minutes |
| Mean-Variance Portfolio Optimization | Markowitz Model, Efficient Frontier, Risk Parity | 20 minutes |
After Completing This Stage
You will be able to:
- Design multi-Agent architecture: Meta Agent, Expert Agents, Risk Agent
- Implement Regime Detection: identify trending, ranging, and crisis markets
- Understand five patterns of Regime misjudgment and response strategies
- Understand the correct use of LLM (enhancement, not replacement)
- Build multi-layer risk control systems with Risk Agent veto power
- Design the portfolio layer: Position Sizing, factor exposure monitoring
- Design online learning and strategy elimination mechanisms
Next Stage
→ Part 5: Production and Practice - Deploy to real environments