AutoCrew is an automated system designed to streamline and govern the end-to-end process of financial model development. Leveraging Large Language Models (LLMs) orchestrated by the CrewAI framework, AutoCrew takes a natural language task description (e.g., "Build a credit risk model") and autonomously executes a multi-step workflow, culminating in a fully validated and audited model.
The system is built around two specialized, sequential Agent Crews to ensure regulatory compliance and robust Model Risk Management (MRM).
- Autonomous Workflow: Executes the entire modeling pipeline from data preparation to final risk sign-off without human intervention.
- Dual-Crew Architecture: Enforces separation of duties via a Modeling Crew (construction) and an MRM Crew (validation and audit).
- LLM Acceleration: Utilizes high-performance LLMs (Meta Llama 3.1 8B via Groq LPU and Google Gemini 1.5 Flash).
- Built-in Simulator: Includes a Stress Testing Engineer for economic shock simulations to validate model stability.
- Compliance Ready: Generates an audit trail through the Compliance Officer.
AutoCrew employs a hierarchical agent system managed by CrewAI.
| Crew | Agents Included | Primary Responsibility |
|---|---|---|
| Modeling Crew | Data Analyst, ML Engineer, Technical Writer | Data cleaning, feature engineering, model training, documentation |
| MRM Crew | Compliance Officer, Stress Testing Engineer, Chief Risk Officer | Code audit, stress testing simulation, regulatory approval |
- Python 3.10+
- Git
git clone https://github.com/YourUsername/AutoCrew-Financial-Agent-System.git
cd AutoCrew-Financial-Agent-System
python -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate
pip install -r requirements.txt
Create a file named .env in the project root (never commit it to GitHub):
GROQ_API_KEY=YOUR_GROQ_API_KEY_HERE
GOOGLE_API_KEY=YOUR_GEMINI_API_KEY_HERE
Execute the main script. You will be prompted to select an LLM (Groq recommended):
python -m src.main
Upon running, AutoCrew processes the Credit Default Prediction dataset, producing a trained model and a detailed report approved by the Chief Risk Officer.
- API keys are protected by the
.gitignore. - All agent actions are logged for auditability, meeting Model Risk Management standards.
Sarim Shah — sarimoman@gmail.com