Agent templateData

ML Engineer AI Agent (Template)

Name: Fleet
Author: Fleet

An ML engineer agent handles the engineering side of machine learning: feature pipelines, model serving infrastructure, evaluation harnesses, and the code that connects trained models to production systems. It is distinct from the research or modeling role; it focuses on production reliability and reproducibility.

ML engineering is highly stack-specific. Whether you use MLflow, Weights and Biases, or a custom tracker; whether you serve with TorchServe, Triton, or a FastAPI wrapper — the agent needs to know your stack to produce usable code. The role-specific prompt is where that context lives.

What this agent owns

Build and maintain feature pipelines that feed model training and inference
Write model serving code, endpoint definitions, and input validation logic
Implement evaluation harnesses that run on each model version before promotion
Write and maintain monitoring code that detects drift or performance degradation and surfaces it through the project alerting pipeline
Manage experiment tracking and model versioning in the project's chosen registry

Recommended model: Claude Sonnet

ML engineering is largely structured pipeline and serving code — feature pipelines, drift checks, canary deployments — which Sonnet handles accurately at lower cost. Use Opus selectively for model-evaluation methodology calls.

Example tasks

Write a feature pipeline that computes rolling aggregates for a recommendation model
Add a data drift check to a model evaluation job using population stability index
Implement a model canary deployment that routes 10% of traffic to the new version
Build an evaluation harness that compares two model versions on a held-out dataset

# create an agent from this template, then start it

$ fleet agent create --name ml-engineer--vendor claude-code --template <template-name>

$ fleet agent start ml-engineer

Find the exact template name with fleet template list.

Related agent roles

Data Engineer →Backend Developer →Performance Engineer →

Run this agent in your fleet

One binary. Five minutes. See every agent, coordinate every handoff, and keep a full audit trail of what your fleet did.

See how it works Install Fleet