ReAct Quant Research Agent

A ReAct-style quantitative research agent capable of:

• Workflow planning
• Market data retrieval
• Momentum factor analysis
• Ranking generation
• Natural-language research explanation

The system follows a layered Agent architecture with:

• Planner layer
• Runtime execution layer
• Pipeline validation layer

---

Architecture

User Request
    ↓
Planner
    ↓
Workflow Construction
    ↓
Runtime Execution
    ↓
Factor Computation
    ↓
Ranking
    ↓
Research Explanation

---

Key Components

Runtime Market Data Step

Implements grouped market-bar retrieval using BaoStock for A-share symbols.

Planner Catalog

Provides planner-visible workflow descriptors for:

• Market data retrieval
• Factor computation
• Ranking
• Explanation generation

Pipeline Validation

Validates:

• Unique step IDs
• Graph connectivity
• Required workflow nodes

ReAct Loop Runtime

Supports:

• Iterative tool execution
• Repair prompts
• Idle detection
• Workflow export

---

Example Workflow

trigger.manual
    ↓
data.market_bars
    ↓
factor.momentum
    ↓
factor.rank
    ↓
research_chat

---

Design Notes

The project explores:

• Tool-driven workflow planning
• Structured Agent execution
• Pipeline validation
• Quantitative research automation