ML Research Expert: Published First Author (Benchmark Authoring)

What you'll do

You write the tasks that frontier models are tested on. AfterQuery wants ML researchers who write code and publish results to design realistic research problems in their own area, author expert-level reference solutions, and write the grading rubrics that decide whether a model's answer is any good.

The ad lists twelve task areas: language models, deep learning, reinforcement learning, vision and generation, robotics, ML systems and efficient ML, optimization and theory, classical and adaptive learning, time series and forecasting, structured and causal reasoning, trustworthy learning, and AI for science. It asks for depth in one or more of them rather than broad familiarity.

It says plainly that this is not annotation or data-labelling work.

Who fits

• At least one first-author research paper in machine learning or a closely related field (required)
• A Master's or PhD in ML, computer science, statistics, mathematics or a related quantitative field; in progress counts, including current PhD candidates
• Hands-on experience writing ML code, whether training, evaluation or systems work
• 1+ years of ML research, with no upper limit

Preferred: several first-author papers, papers reporting measured gains over a baseline, and publications at venues such as NeurIPS, ICML, ICLR, CVPR, ACL, EMNLP, CoRL or MLSys.

The application form asks for a link to your first-author paper and the headline result from it, both required. The ad says candidates who meet the requirements and supply those materials are prioritised for review, and that AfterQuery runs thorough background checks.

What it pays

The board shows $150/hour. The ad describes the hours as fully remote, asynchronous and flexible, between 5 and 40 a week. It does not publish payment terms such as the payout schedule or a minimum weekly commitment; the AfterQuery company page covers how the company pays contractors in general.

Worth knowing

Good:

• One of the higher stated rates among AfterQuery's engineering listings
• Work that uses research skill rather than labelling speed
• Flexible hours, and a wide range of task areas to match your specialty

Less good:

• A published first-author paper is a hard gate, and the form asks you to prove it up front
• Background checks are part of the process
• The ad does not describe the assessment, so expect the one-attempt unpaid assessment the company page describes
• No guaranteed weekly volume is stated

About this listing

Read from the AfterQuery Experts board on 18 October 2026; the listing was posted on 15 September 2026 under the Software Engineering department. See AfterQuery.