Back-End Engineer

AI Trainer Jobs; United States; Remote; USD 69,000 - 165,000; Full time

Job summary

AI Trainer Jobs is seeking a Back-End Engineer for a remote engineering review track. You will evaluate production code, debug traces, and assess developer-facing AI outputs against real-world correctness standards.

Responsibilities include reproducing failures, writing the unit tests the model should have written, and explaining the fixes so the modeling team can target the gap. This is a remote contractor role, US-eligible.

Qualifications

• Open-source contributions or a public portfolio that demonstrates production-quality code.
• Experience with the target language's standard tooling, linting, and testing workflows.

Responsibilities

• Run and reproduce candidate code outputs in a sandboxed environment for Back-End Engineer assignments.
• Grade backend engineering solutions for correctness, style, and edge-case handling.
• Write minimal failing tests that demonstrate the bug a model output missed.
• Compare paired solutions and rank them with a written rationale tied to the rubric.
• Tag failure modes (compile error, runtime crash, off-by-one, security issue) with severity scores.

Skills

Engineering experience

Unit tests

Testing framework

Written reasoning

Async availability

Code review experience

Tools

pytest

Jest

JUnit

Go test

RSpec

Job description

Role details

Back-End Engineer is a remote engineering review track for evaluating production code, debugging traces, and developer-facing AI outputs against real-world correctness standards. Reviewers reproduce failures, write the unit test the model should have written, and explain the fix so the modeling team can target the gap.

Category: Coding, SWE & Agent Evaluation · Pay: $50-$120 / hr · Location: Remote - US-eligible · Contractor

About the role

Back-End Engineer is a remote engineering review track for evaluating production code, debugging traces, and developer-facing AI outputs against real-world correctness standards. Reviewers reproduce failures, write the unit test the model should have written, and explain the fix so the modeling team can target the gap.
Engineering model quality lives or dies on whether the generated code actually compiles, passes tests, and handles edge cases. AuraOne pairs experienced engineers with the modeling team to grade outputs the way a code reviewer would.
Judge generated code and software engineering agents. Read their debugging traces.

Responsibilities

• Run and reproduce candidate code outputs in a sandboxed environment for Back-End Engineer assignments.
• Grade backend engineering solutions for correctness, style, and edge-case handling.
• Write minimal failing tests that demonstrate the bug a model output missed.
• Compare paired solutions and rank them with a written rationale tied to the rubric.
• Tag failure modes (compile error, runtime crash, off-by-one, security issue) with severity scores.

What you should bring

• Strong day-job engineering experience - you can read, run, and debug unfamiliar code for Back-End Engineer work.
• Comfort writing concise unit tests that capture a single failure mode.
• Familiarity with a testing framework. pytest, Jest, or JUnit. Go test, RSpec, or whatever you use.
• Clear written reasoning — your review note has to convince another senior engineer.
• Reliable async availability for at least 10 hours per week.
• Prior code-review or technical-interview-grading experience is a plus.

Role signals

Example tasks

• Reproduce a generated engineering solution to a coding task, run the test suite, and grade it.
• Write the smallest failing test that demonstrates a model's edge-case bug.
• Compare two paired solutions and rank them with a written rationale tied to the rubric.
• Triage a security issue surfaced by a model and document the patch the model should have produced.

Useful experience

• Open-source contributions or a public portfolio that demonstrates production-quality code.
• Experience with the target language's standard tooling, lint<|reserved_200016|><|start|>assistant<|channel|>analysis<|message|>We need to produce final output JSON with FinalAction, FormattedDescription, IsExpired. We have to evaluate whether the content should be refined or not based on developer instruction. Let's examine the input. Input is a job description with placeholders like