Mindrift

Freelance Agent Evaluation Engineer

Mindrift
Apply
18 hours ago
Remote, Brazil +3 moreSenior

Responsibilities

  • Build realistic simulated developer environments containing codebases, infrastructure, tickets, documentation, and conversation history.
  • Design tasks from intermediate environment states, including prompts and criteria defining successful solutions.
  • Write functional and integration tests that accept valid agent solutions and reject incorrect ones.
  • Review agent solutions, analyze failures, and iterate on tasks and tests based on QA feedback.
  • Create evaluations that reveal meaningful differences in AI coding-agent performance.

Requirements

  • At least 5 years of software development experience.
  • A master's degree in Computer Science, Software Engineering, Data Science/Data Analytics, Artificial Intelligence/Machine Learning, Computational Linguistics/Natural Language Processing, Information Systems, or a related field; a bachelor's degree is accepted with 5 years of field experience.
  • At least 3 years of professional experience in related roles or domains, specifically for QA-automation/testing or cybersecurity roles.
  • Experience with Python and FastAPI, JavaScript/TypeScript and React, Docker, Postgres, Kafka, and Redis.
  • Experience writing functional and integration tests.
  • English proficiency at B2 level or higher.
  • Submit a CV in English and indicate English proficiency.

Benefits

  • Part-time, remote, freelance project-based work that can fit around other professional or academic commitments.
  • Work on advanced AI projects and build portfolio experience influencing future AI systems.
  • Paid per accepted task, with rates determined by qualification tier and completion efficiency.
  • Effective compensation can reach up to $30 per hour equivalent.
Mindrift

About Mindrift

1,001-5,000 employees

Mindrift builds an expert-sourcing platform that connects domain specialists to project-based work training and evaluating generative AI models, including supervised fine-tuning, RLHF, evaluation, and red-teaming. It is built and operated by Toloka, part of Nebius Group, and run from Amsterdam, Netherlands. Work is fully remote and freelance, serving global technology companies developing and improving large AI systems.

Contact me