Software Engineer - AI Model Evaluation & Coding Tasks

We assess engineers on the work they have already done, not on interviews.

Cookiy AI connects working software engineers with AI labs and data buyers who need two things: real agent coding sessions, and the engineers behind them for ongoing evaluation and coding work.

We are opening a part-time, remote, contract engineering bench for our first enterprise client, an internationally recognised technology company. Work is project-based and paid hourly.

Work from anywhere in the world

You can be based in any country, any city, any time zone. There is no location requirement of any kind. The Singapore location on this listing is only our company's registered base, which LinkedIn requires us to enter. You do not need to be in Singapore, and you do not need to be in any particular region. All work is asynchronous, so you choose the hours that suit your time zone.

What the work is

• Evaluating AI model output on real engineering tasks: correctness, code quality, reasoning, tool use
• Writing and reviewing reference solutions for coding problems
• Rubric-based scoring and structured annotation of agent trajectories
• Short scoped build tasks defined by the client

Remote, asynchronous, part-time. You keep your day job. Typical commitment is 5 to 15 hours per week, scheduled entirely by you.

How our process works, and why it is different

Every other company in this space starts with an interview or a take-home test you complete for free. We do not.

Step 1. License your past agent sessions. Install our open source skill, review what it finds on your machine, and choose which of your Claude Code or Codex sessions you are willing to license. Secrets and personal data are scrubbed locally on your own machine, and nothing leaves your laptop until you approve it.

You are paid for this whether or not we ever offer you project work. Sessions earn each time a buyer licenses them. This is not an unpaid application exercise. It is a transaction that stands on its own.

Step 2. Assessment. Buyers evaluate the licensed sessions. Your planning, your debugging, your corrections to the model and your test discipline are the signal. A strong engineer is visible in a month of real work in a way they never are in a 45 minute call.

Step 3. Matching. Engineers whose sessions show the depth our client is looking for are invited onto the paid bench and matched to tasks that fit their actual stack.

Be clear on this: licensing your sessions does not guarantee project work. Many engineers will license sessions, earn from them, and not be matched, especially early while we are staffing a single client. The licensing income is yours either way. We would rather say this plainly than have you find out later.

How to apply. Read this carefully.

We do not accept one click applications. To be considered you must complete all three steps below. Applications that skip them are not reviewed.

1. Install the open source skill at github.com/cookiy-ai/sell-sessions-skill. The README gives you three ways to do it. The fastest is one command in your terminal.
2. Upload your sessions for licensing and assessment. You choose exactly which ones.
3. Find the survey code in the bottom right corner of the local app, and enter it in the screening question on this application. Without the correct code we cannot verify that you completed the process, and your application will not be reviewed.

This takes about ten minutes. We ask for it because our client needs engineers who read instructions closely and follow them precisely, and because the sessions themselves are how we assess you.

What you must not upload

Only license sessions from work you own or are permitted to share: personal projects, open source, your own products, or freelance work where your client has agreed.

Do not upload sessions containing your employer's proprietary code. Check your employment agreement first. We reject sessions we believe are not yours to license. Our buyers require verifiable provenance, and so do we.

What we are looking for

• Three or more years building and shipping production software
• Current, daily use of an AI coding agent such as Claude Code, Codex or Cursor
• Real depth in at least one area: backend, infrastructure, data, systems, frontend or ML
• Written English strong enough for precise technical judgments
• Recent agent sessions that you are legally free to license

Independent contractor engagement, paid in USD. Open to engineers everywhere.