Custom Agent for Analysis of FRED Data

A demo that uses a Council Agent to access data from FRED and generate Python code for data analytics. Runnable as a notebook or as a Flask/JavaScript app with a UI.

There is a custom controller, which is a development version of LLMInstructController

• src/agent/council_controller.py

Prerequisites

To run the demo as-is, you'll need an OpenAI API Key and a FRED API Key.

Council Components

We define an Agent:

graph TB
    User[User] -->|Messages| LLMInstructController
    subgraph Agent[Agent]
        LLMInstructController[LLMInstructController]
        BasicEvaluator[BasicEvaluator]
        subgraph Chains
            Chain1[fred_data_specialist]
            Chain2[data_analysis_code_editing_and_execution]
            Chain3[data_analysis_code_editing]
            Chain4[code_execution_and_correction]
            Chain5[general]
        end
        LLMInstructController -->|"chain;score;instructions"| Chains
        Chains --> BasicEvaluator
        BasicEvaluator --> LLMInstructController
    end

Loading

With Chains:

graph TB
    subgraph B[Chain: fred_data_specialist]
        G1[Skill: FredDataSpecialist] --> I1[Skill: ParsePythonSkill]
    end

Loading

graph TB
    subgraph C[Chain: data_analysis_code_editing_and_execution]
        H2[Skill: PythonCodeEditorSkill] --> I2[Skill: ParsePythonSkill] --> J2[Skill: PythonExecutionSkill]
    end

Loading

graph TB
    subgraph D[Chain: data_analysis_code_editing]
        H3[Skill: PythonCodeEditorSkill] --> I3[Skill: ParsePythonSkill]
    end

Loading

graph TB
    subgraph E[Chain: code_execution_and_correction]
        I4[Skill: ParsePythonSkill] --> J4[Skill: PythonExecutionSkill]
    end

Loading

graph TB
    subgraph F[Chain: general]
        K[Skill: GeneralSkill]
    end

Loading

Running the demo

• Create a new Python/conda environment
• Install dependencies using pip install -r requirements.txt (outdated - this should be much smaller)
• Set up a the Python sandbox
  • cd to src/agent
  • python -m venv code_sandbox
  • source code_sandbox/bin/activate
  • pip install fredapi pandas plotly seaborn scikit-learn
• Populate your .env file with
  • Make a copy of .env.example and rename it to .env
  • OpenAI API Key
  • FRED API Key
  • Python sandbox bin directory
• Run the notebook src/run_agent.ipynb
• Run the Flask app
  • cd src/flask-app
  • python app.py and open the webpage (from Finder/Explorer etc.) src/flask-app/index.html

Note

If you are experiencing problems with the Flask app / UI, try restarting both the Flask app and your browswer.