Research Engineer - Code Generation & Model Evaluation

What you'll do

Despite the "Research Engineer" title, this is hands-on coding work, not a research post. You work inside real codebases (Python 3, Java, Rust, C++, Go or TypeScript) and the output is used to train and evaluate code-generation models.

The ad lists:

• Debugging and resolving issues across diverse codebases
• Designing and implementing features in code-generation workflows
• Refactoring and optimising for performance and maintainability
• Writing and reviewing real-world coding tasks that measure AI model performance
• Writing clear feedback and annotations for model training and assessment
• Working with other open-source contributors and documenting decisions

In practice this sits between a software engineering contract and benchmark authoring: you need to be good enough to fix the bug yourself, and disciplined enough to write the task so a model's attempt can be graded unambiguously.

Who fits

• Expertise in competitive programming and coding problem analysis
• Strength in at least one of Python 3, Java, Rust, C++, Go or TypeScript, with solid algorithms and data structures
• A record of open-source contributions or collaborative software projects
• Careful reading of problem constraints and alternative solution paths
• Thorough validation of code and output consistency
• Clear written communication, and comfort working independently to tight deadlines

These are listed as required, unlike most micro1 ads. The competitive-programming plus open-source combination is the distinctive bar; strong day-job engineers without either may find it harder.

What it pays

The band is $50–100/hour, but the ad is explicit that pay is output-based: you are paid per task that meets the project specification, time per task varies with your experience and workflow, and there is a minimum number of tasks per week (the figure is not published).

So the hourly band is an effective-rate expectation rather than a guarantee. Fast, accurate contributors will land higher; tasks that fail review earn nothing.

Start

micro1 says it typically fills roles like this within 48 hours and expects first tasks within 24–48 hours of finishing onboarding. If you apply, be ready to start straight away.

Worth knowing

Good:

• 100 openings and a fast start
• Six languages qualify, so you can work in your strongest one
• Pay mechanics are stated rather than left to guesswork
• No AI experience needed, and no country restriction stated

Less good:

• Per-task pay: rejected work is unpaid and the effective rate depends on your speed
• A weekly task minimum applies, but the number is not published
• Competitive programming and open-source history are both asked for
• "Tight deadlines" is in the ad for a reason
• Contractor engagement: no benefits, no notice period, your own tax to manage

About this listing

Posted by micro1 on its own jobs portal and read on 24 September 2026. The band, the 100 openings, the languages, the output-based pay terms and the start timeline are from the listing. See micro1.