ADR-001: AI Base Template

Status

Accepted

Context

A minimal Python template for AI/ML projects that provides a standardized starting point with modern tooling, best practices, and pre-configured libraries for rapid prototyping and clean architecture.

Decision

Build a lightweight, opinionated template that balances simplicity with completeness, providing essential ML/data science tools while maintaining flexibility for different project types.

Consequences

• Pros: Faster project initialization, consistent structure across projects, reduced setup time, built-in best practices
• Cons: May include unnecessary dependencies for simple projects, opinionated choices may not suit all use cases

Technical Specification

• Stack: Python 3.12, FastAPI, Pydantic, PyTorch, scikit-learn, XGBoost/LightGBM
• API: FastAPI-based REST API with automatic OpenAPI documentation
• Dependencies: uv for package management, pre-configured ML/data science libraries
• Data Flow: Input → FastAPI endpoints → Processing modules → Response models
• State Management: Stateless by default, configurable via environment variables
• Scaling: Horizontal scaling supported via FastAPI async capabilities

Integration Points

• Consumes: Environment variables, .env files for configuration
• Provides: REST API endpoints, Jupyter notebook support for experimentation
• Protocols: HTTP/REST, async support via FastAPI

Non-Functional Requirements

• Performance: Async request handling, efficient ML model serving
• Availability: Development/research focused, no specific SLA
• Security: Environment-based configuration, no hardcoded secrets
• Scalability: Suitable for prototypes to small/medium production deployments

Deployment

• Platform: Local development, containerizable for K8s/cloud deployment
• Configuration: Environment variables via .env files, python-dotenv
• Resources: Depends on ML model complexity, baseline ~1GB RAM, 1 CPU core