Join the roster
Your expertise builds better AI.
Join a vetted network of engineers, researchers, and domain experts doing flexible, remote, well-paid work evaluating and training frontier models.
Apply NowOpen opportunities
We're hiring coding experts who train AI models.
One role, clearly defined. If you match the profile below, we want to hear from you.
AI Training Reviewer
Evaluate and write tasks that train frontier coding agents. Work runs in containerised terminal environments — you'll author scenarios, review agent transcripts, and produce RLHF and SFT data that model teams actually use.
What we look for
Five things that matter for this work.
These aren't nice-to-haves. They reflect what terminal-bench-style evaluation and training data work actually requires day to day.
Coding proficiency
Proficient in JavaScript, Python, or Java. Able to read, write, debug, and review production-quality code across real engineering contexts, not just solve isolated algorithm puzzles.
Docker fluency
Comfortable operating in containerized environments: building images, debugging container setups, and reasoning about dependency isolation. Our tasks run in Docker-based terminal environments to guarantee reproducibility.
Scenario-based reasoning
Strong judgment applied to open-ended engineering problems. You can diagnose whether an agent failed because of a genuine model limitation or a poorly specified task — the kind of call terminal-bench-style review depends on.
RLHF and SFT expertise
Hands-on understanding of reinforcement learning from human feedback and supervised fine-tuning pipelines. The data you produce needs to be usable by a model team, not just technically correct in isolation.
AI domain fluency
Broad understanding of how modern LLMs and coding agents are trained, evaluated, and deployed. You understand why a task or eval matters, not just how to complete it.
Why experts choose Caudal
Serious work. Transparent terms. Your schedule.
Transparent compensation
Rates are structured around task complexity and domain expertise. Payment details are shared clearly before you take on any work — and paid reliably on a consistent cadence.
Fully flexible
Remote, async, no minimum commitment. Most contributors work part-time alongside their primary job or research. You choose your projects and your pace.
Work that matters
Your assessments directly shape how frontier models reason and behave. Work alongside other top practitioners, with real reviewer feedback on every submission.
How it works
From sign-up to paid work in four steps.
Sign up
Create your account and tell us about your domain background. Takes under two minutes.
Complete the assessment
A 15-minute evaluation covering software engineering, CLI tooling, and code analysis — the actual work you'll do. Auto-scored and manually reviewed.
Get matched
Our team reviews your assessment and matches you to projects in your domain, usually within 48 hours.
Start working and get paid
Take on tasks at your own pace, submit your work for review, and get paid per accepted task.
What contributors say
From the people doing the work.
"The tasks are genuinely hard — debugging agentic pipelines, evaluating model outputs against rubrics. It's closer to senior engineering review than annotation work."
Arjun Nair
Staff Engineer, Backend Systems · ex-Flipkart
"I get real feedback on every submission, not just a pass/fail. That alone makes it worth it. The problems are interesting and the pay is consistent."
Priya Venkataraman
SDE-2, Distributed Systems · Amazon India
"I've done RLHF work elsewhere, but the rubric quality here is a different level. You can tell the people writing these tasks actually understand the models."
Karan Mehta
CS grad, ML Systems · IIT Bombay
Frequently Asked Questions
- How does the screening process work?
- After signing up you complete a short assessment covering software engineering, CLI tooling, and code evaluation — the areas we work in. General assessment is auto-scored; scenario based assessment responses are reviewed by our team. We typically share results within 48 hours.
- How much can I earn?
- Rates are structured around task complexity and domain expertise, not volume throughput. We don't publicise specific figures here, but payment details are shared clearly before you take on any work.
- Is this full-time work or flexible?
- Entirely flexible. Most contributors work part-time, fitting tasks around their primary job or research. There is no minimum commitment or fixed schedule.
- What kinds of tasks will I work on?
- Evaluating model outputs, writing scoring rubrics, solving engineering problems, reviewing AI-generated code, and constructing adversarial test cases for software benchmarks. Every task is matched to your coding background.
- How long before I get my first project?
- Once your assessment is reviewed and approved, we typically match contributors to their first project within 48 hours.
- Can I work from anywhere?
- Yes. Caudal is fully remote. Contributors work from wherever they are, with no geographic restrictions.
- How and when do I get paid?
- Payment is issued after task review on a regular cadence. We support bank transfer and common digital payment methods. Full details are provided during the intake process.
Get new roles in your inbox.
Not ready to apply yet? Drop your email and we'll let you know when new opportunities open up in your domain.
Ready to contribute
Your expertise belongs on the frontier.
Join 2,000+ vetted engineers, researchers, and domain experts already building the data behind frontier AI.
Apply Now