How biyihan2005/alpha_research_agent works
Alpha Research Agent structures quantitative factor research from inspiration and hypotheses through constrained formula validation, local data evaluation, experiment memory, and reporting. The graph preserves the documented provider, DSL, evaluation-metric, and optional publishing capabilities. Relationships are based on the README workflow and shown source references; unsampled implementation details are not treated as verified call paths.
Main entry points: base.py LLM Interface · retriever.py Idea Retriever · hypothesis_agent.py Hypothesis Agent · formula_agent.py Formula Agent
Components 20
Research Ideation
- Idea Retriever
alpha_agent/inspiration/retriever.py - Idea Cards
alpha_agent/inspiration/idea_card.py - Hypothesis Agent
alpha_agent/agents/hypothesis_agent.py - Formula Agent
alpha_agent/agents/formula_agent.py - Critique Agent
alpha_agent/agents/critique_agent.py
Formula and Data
- DSL Validator
alpha_agent/dsl/validator.py - Formula Parser
alpha_agent/dsl/parser.py - Dataset Catalog
alpha_agent/datasets/catalog.py - Data Providers
alpha_agent/datasets/providers - Panel Loader
alpha_agent/data/loader.py
Evaluation and Memory
- Factor Evaluator
alpha_agent/evaluation/evaluator.py - Quality Gate
alpha_agent/evaluation/quality_gate.py - Experiment Memory
alpha_agent/memory/factor_store.py
Reporting and Integration
- Research Report
alpha_agent/agents/report_agent.py - Feishu CLI
alpha_agent/integrations/feishu_cli.py
LLM Clients
- LLM Interface
alpha_agent/llm/base.py - Mock LLM
alpha_agent/llm/mock_client.py - DeepSeek Client
alpha_agent/llm/deepseek_client.py
Other components
- Researcher
- DeepSeek API
Connections 20
- Researcher → Idea Retriever: provides goal
README.md - Idea Retriever → Idea Cards: builds cards
alpha_agent/inspiration/retriever.py - Idea Retriever → Hypothesis Agent: informs
README.md - Hypothesis Agent → Formula Agent: guides formulas
README.md - Hypothesis Agent → LLM Interface: generates with
alpha_agent/agents/hypothesis_agent.py - Formula Agent → LLM Interface: generates with
alpha_agent/agents/formula_agent.py - Critique Agent → LLM Interface: generates with
alpha_agent/agents/critique_agent.py - Mock LLM → LLM Interface: implements
alpha_agent/llm/mock_client.py - DeepSeek Client → LLM Interface: implements
alpha_agent/llm/deepseek_client.py - DeepSeek Client → DeepSeek API: requests completions
alpha_agent/llm/deepseek_client.py - Formula Agent → DSL Validator: submits formulas
README.md - DSL Validator → Formula Parser: validates before evaluation
alpha_agent/dsl/parser.py - Dataset Catalog → Data Providers: registers providers
alpha_agent/datasets/catalog.py - Data Providers → Panel Loader: normalizes data
alpha_agent/datasets/providers/sample_provider.py - Data Providers → Factor Evaluator: supplies panel
README.md - DSL Validator → Factor Evaluator: passes valid formulas
README.md - Factor Evaluator → Quality Gate: submits metrics
README.md - Quality Gate → Experiment Memory: stores accepted results
README.md - Experiment Memory → Research Report: provides experiment records
README.md - Research Report → Feishu CLI: publishes optionally
README.md