ai-ml-engineer_skill

Overview

This skill is a Copilot-style AI/ML Engineer that assists end-to-end with machine learning projects: model design, training, evaluation, deployment, and MLOps. It guides you through data preparation, model selection, hyperparameter tuning, and productionization while recommending frameworks and tools. Use it to accelerate experiments, produce reproducible pipelines, and implement monitoring and CI/CD for models.

How this skill works

The agent inspects project goals, dataset characteristics, and constraints to recommend an actionable plan: data preprocessing steps, model architectures, training loops, evaluation metrics, and deployment patterns. It generates code snippets, configuration suggestions, and stepwise deliverables (data classes, trainers, inference APIs) and proposes MLOps setups (versioning, monitoring, CI/CD). It enforces a phased, single-question workflow to collect requirements before implementation.

When to use it

• Starting a new ML/AI project and needing an implementation plan
• Improving or debugging existing models (performance, bias, data issues)
• Preparing models for production: containerization, APIs, and monitoring
• Designing MLOps pipelines: model versioning, automated training, drift detection
• Selecting architectures and tooling for NLP, CV, time series, or tabular tasks

Best practices

• Collect one clear requirement at a time and confirm before proceeding
• Create English documentation first, then provide a required Japanese translation
• Always check project steering/architecture guides if present to align choices
• Use small reproducible artifacts: dataset class, trainer, inference API, and tests
• Adopt model versioning and monitoring from the start to prevent regressions

Example use cases

• Designing and training an image classification pipeline with PyTorch and TorchServe
• Fine-tuning a transformer for domain-specific text classification using Hugging Face
• Building a production REST API (FastAPI + Docker + Kubernetes) for model serving
• Implementing CI/CD for automated retraining and deployment with MLflow and GitHub Actions
• Setting up monitoring for model drift and performance alerts with Weights & Biases