MLE Bench: ML Engineers

What you'll do

MLE Bench is a benchmark of real machine learning engineering work, and this role supplies the engineering behind evaluation tasks of that kind. You work in production-grade ML codebases: building, running and changing training, evaluation and inference pipelines; preparing datasets, features and metrics; and debugging and refactoring systems for correctness and performance. You also study model behavior, failure modes and edge cases, write reproducible, documented Python, take part in code reviews and work with researchers to design hard, realistic ML engineering tasks that can test AI systems.

Who qualifies

• At least 3 years as an ML engineer or ML-focused software engineer
• Strong Python for ML and data workflows
• Hands-on experience with training, evaluation and inference pipelines
• Solid ML fundamentals (supervised and unsupervised learning, evaluation metrics, optimization)
• Experience with PyTorch, TensorFlow, JAX or similar
• The ability to navigate and modify complex real-world ML codebases and write maintainable code
• Excellent spoken and written English

What it pays

The ad does not publish a rate. Turing sets rates per project; see Turing.

Worth knowing

Good:

• Engineering work on realistic ML systems rather than labeling or rating
• Clear terms: at least 4 hours a day and a minimum of 20 hours a week, with 4 hours of overlap with PST; 3 months, adjustable based on engagement
• The vetting is stated: a 60-minute technical interview with a live coding challenge

Less good:

• Open only to residents of India, Pakistan, Nigeria, Kenya, Egypt, Ghana, Bangladesh, Turkey, Brazil and Mexico
• Contractor assignment with no medical or paid leave
• No pay figure, so the rate is unknown until matched
• A 20-hour weekly minimum plus PST overlap is a real schedule constraint

Turing also lists a Data Analyst version of this project, published separately.

About this listing

Read from Turing's job board on 18 October 2026; the role was first posted in January 2026. Requirements and terms are Turing's; the comparison and caveats are ours. See Turing.