AfterQuery contracts data scientists and ML engineers with 1–3 years of professional experience to build, train, and document models that become training and evaluation material for AI systems. Work is remote and asynchronous over a 2–3 week window, at roughly 10–20 hours per week on the projects you accept.

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

You take on self-contained ML and analytics tasks that mirror what you would do in a working data science role: framing a prediction problem, cleaning and engineering features, training supervised or unsupervised models, tuning hyperparameters, and reporting what actually moved the metric. The deliverable is rarely just a notebook that runs — it is the reasoning trail. Expect to write down your assumptions, your validation scheme, why you rejected the approaches you rejected, and where the model would fail in production. That written record is the point: it is what makes the task usable as training or evaluation data for models being taught to do quantitative work.

Tasks arrive project-by-project rather than as a steady queue. Some will be model-building end to end; others will be narrower — diagnose a leakage problem, redo an evaluation that was set up wrong, or produce a clean baseline against which something else is measured. Datasets vary in size and messiness, and tabular work sits alongside text, time series, and occasional deep learning depending on the project.

What the screen looks for

AfterQuery's screening is credential-gated at the front — a master's in a quantitative field and at least one first-author publication or first-author-equivalent work are stated requirements — and then goes technical. Expect follow-ups that push past vocabulary: not "what is regularization" but "what did you do when your cross-validation score and your holdout score disagreed." Screeners probe whether you can defend a validation design, explain a metric choice against an alternative, and describe a modelling decision you got wrong. Vague or textbook answers are the main failure mode; specific, slightly unflattering stories from real projects are the strongest.

Logistics

Fully remote, fully asynchronous — no fixed hours or standing calls
10–20 hours/week expected on projects you accept, over a 2–3 week engagement window
Project-based contractor pay, observed in the $60–100/hr range, with the top of the band tied to task complexity and track record
You supply your own machine and Python/ML environment
TA experience in ML, data science, or applied statistics is a stated preference, not a requirement

Machine Learning / Data Science Expert

What the work involves

What the screen looks for

Logistics

Interview prep — 10 questions