Skip to content

README.md

Latest commit

 

History

History
137 lines (99 loc) · 4.38 KB

README.md

File metadata and controls

137 lines (99 loc) · 4.38 KB

Quant Apprentice 📈

Quant Apprentice is an autonomous AI agent for sophisticated, end-to-end financial analysis. This repository showcases the evolution of the agent through three distinct versions, from a simple rule-based system to a complex, self-reflecting agentic workflow powered by Google's Gemini Pro.

Project Versions

This repository is organized into three distinct, self-contained versions to demonstrate the project's development lifecycle. Please see the entry point notebook inside each folder for specific instructions and code.

📁 v0_no_llm/

  • Description: A foundational version of the agent built with pure Python, without any Large Language Models. It uses hard-coded rules for analysis and demonstrates the basic structure of an automated research agent.
  • How to Run: See the demo.ipynb notebook inside this directory.

📁 v1_llm_linear/

  • Description: The first integration of an LLM (Gemini) into the workflow. This version replaces the rule-based logic with LLM prompts for tasks like news analysis and report synthesis, but still follows a linear, top-to-bottom script.
  • How to Run: See the v1_linear_agent.ipynb notebook inside this directory.

📁 v2_llm_graph/

  • Description: The final and most advanced version of the agent, rebuilt as a stateful, cyclical graph. This version uses LangGraph, incorporates a vector database for semantic memory (RAG), an expanded toolset (SEC filings), and a robust self-critique loop. This is the final version for submission.
  • How to Run: See the v2_graph_agent.ipynb notebook inside this directory.

🧪 Testing

The project maintains comprehensive test coverage across all components:

# Run full test suite with coverage reporting
python -m pytest tests/ --cov=v2_llm_graph --cov-report=term-missing -v

# Run only unit tests (excluding integration tests)
python -m pytest tests/ -v -k "not integration"

Current test coverage:

  • Overall coverage: 95%
  • Core components:
    • Agent Graph: 90%
    • Tools (News, SEC, Financial): 100%
    • Vector Memory: 100%
    • Workflows: 100%

🚀 Getting Started (for Final Version 2)

Platform Support

This project was developed and tested on macOS but should be platform-independent as it uses:

  • Python's standard library and cross-platform packages
  • Conda for environment management
  • SQLite-based vector storage (ChromaDB)
  • Standard file paths with os.path for compatibility

If you encounter any platform-specific issues, please open an issue on GitHub.

1. Set up Environment

It is recommended to use a Conda environment.

conda env create -f src/environment.yml
conda activate quant-apprentice

2. Create .env File

The agent requires API keys for various data sources. Create a .env file in the root directory and
add the following, replacing your_key_here with your actual API keys:

GOOGLE_API_KEY="your_key_here"
NEWS_API_KEY="your_key_here"
SEC_API_KEY="your_key_here"
FRED_API_KEY="your_key_here"

3. Install Dependencies

Install the required Python packages:

pip install -r requirements.txt

4. Verify Setup

Run the test suite to ensure everything is set up correctly:

# Run tests with coverage report
python -m pytest tests/ --cov=v2_llm_graph --cov-report=term-missing -v

Project Structure (v2)

src/v2_llm_graph/
├── v2_graph_agent.ipynb     # Main entry point
├── src/
│   ├── agent_graph.py       # Core agent architecture
│   ├── tools/               # Data gathering tools
│   │   ├── financial_data_fetcher.py
│   │   ├── news_fetcher.py
│   │   └── sec_filings_fetcher.py
│   ├── workflows/           # Analysis chains
│   │   ├── news_analysis_chain.py
│   │   ├── report_evaluator.py
│   │   └── specialist_router.py
│   └── memory/             # Vector storage
│       └── vector_memory.py
└── tests/                  # Comprehensive test suite
    ├── test_v2_tools.py
    ├── test_v2_memory.py
    ├── test_v2_workflows.py
    └── test_v2_agent_graph.py

Error Handling

The system includes comprehensive error handling for:

  • API failures and timeouts
  • Memory system errors
  • LLM interaction issues
  • Data processing edge cases

All error cases are covered by integration tests to ensure robust operation.