Machine Learning Engineer Talent Network

$70–250/hrrate as listed

Mercor maintains a standing pool of machine learning engineers who are matched to AI lab projects as they open, rather than to one fixed role. Work involves building realistic ML engineering tasks, grading model outputs against them, and writing feedback that explains where a model's reasoning or code breaks down.

Summary, eligibility guidance and screening preparation are prepared by Sidequest from the platform's posting. Check the source listing before applying.

What the work involves

This is an open application to Mercor's ML engineering talent network, not a posting for a single project. Once verified, you're matched on a rolling basis to contracts from AI labs that need people who have actually shipped models. The day-to-day varies by client, but the recurring patterns are: authoring tasks drawn from real ML engineering work (a training loop that silently diverges, a data pipeline with leakage, a deployment that degrades under load), producing reference solutions, and grading model attempts against them. A second common track is comparative evaluation — ranking two or more model responses on correctness, efficiency, and whether the approach would survive contact with production.

Good work here reads like a thorough code review. Labs are not paying for a verdict; they're paying for the reasoning behind it. If a model chooses `BCEWithLogitsLoss` over a manual sigmoid-plus-BCE, you should be able to say why that matters numerically. If it recommends distributed training for a workload that fits on one GPU, you should say so and quantify the cost.

What the platform screens for

Mercor's process is resume upload, credential verification, then an AI-led video interview. The interview probes depth rather than breadth: expect follow-ups that push past your first answer on specific projects you've listed. Vague ownership claims tend to unravel — be ready to name the model architecture, the dataset scale, the metric you moved, and what you tried that didn't work. Screens tend to favor candidates with production experience over coursework or Kaggle-only backgrounds, and they weight PyTorch or TensorFlow fluency, deployment and MLOps exposure, and the ability to explain a technical tradeoff clearly in a few sentences.

Logistics and pay

Fully remote and largely asynchronous; most projects set weekly throughput expectations rather than fixed hours.
Typical commitments observed at 15–30 hours per week, with project durations from a few weeks to several months.
Observed rates on this network span $70–250/hr, set per project and driven by seniority, specialization, and client. This is a reported range, not a guarantee — network membership carries no commitment of work.
Onboarding to a given project usually includes a rubric, calibration examples, and a period where your gradings are spot-checked against other reviewers.

What they look for

Professional experience developing and shipping ML models in production
Strong Python plus deep fluency in PyTorch or TensorFlow
Model deployment and MLOps: serving, monitoring, versioning, CI for ML
Ability to write clear technical critique explaining why an approach fails
Comfort working independently and asynchronously on remote contracts
Familiarity with LLM training or fine-tuning workflows (advantageous, not required)

Before you apply

Would you pass this screen?

Drawn from what this listing actually requires. Answered in your browser — nothing is sent to us, and we do not store your answers.

Do you have at least 2 years of professional experience training, tuning, or deploying machine learning models in a work or research setting?
Professional ML development experience · required
Do you write Python regularly as part of your current or recent work?
Python proficiency · required
Have you built models using PyTorch, TensorFlow, or a similar deep learning framework (not just tutorials—actual projects)?
ML framework hands-on experience · required
Are you able to work independently in a remote environment and commit 15–30 hours per week to contract projects?
Remote work availability · preferred

0 of 4 answered

Included with this listing

Interview prep — 12 questions

Matched to what this platform's screening actually measures. Answer each out loud in 60–90 seconds: claim → concrete example → implication.

What it tests: Depth of practical ML development experience: can you articulate the full pipeline from problem definition through validation, not just talk about frameworks?
A strong answer: A clear narrative covering the business problem or research question, the dataset and preprocessing choices, the specific framework and architecture selected with justification, training procedures (hyperparameter tuning, stopping criteria), and how performance was measured against a baseline or success threshold.
Where people lose it: Vague hand-waving about 'using TensorFlow' or 'building a model' without concrete details on data handling, architecture decisions, or validation methodology—interviewers catch tutorial-level depth instantly.