Python Engineer - Freelance AI Trainer
Berlin, Germany · ಅರೆಕಾಲಿಕ
ಅರ್ಜಿ ಸಲ್ಲಿಸುವವರಲ್ಲಿ ಮೊದಲಿಗರಾಗಿರಿ
- ಅನುಭವ
- 4–5 yrs
- ಸಂಬಳ
- USD 75 / hour
- ತೆರೆಯುವಿಕೆಗಳು
- 1
- ಪೋಸ್ಟ್ ಮಾಡಲಾಗಿದೆ
- 2 ಗಂಟೆಗಳ ಹಿಂದೆ
- ಕೆಲಸದ ಮೋಡ್
- ಕಚೇರಿಯಲ್ಲಿ
- ಅರ್ಹತೆ
- Applicants must provide their CV in English and indicate their English proficiency level. Candidates with a strong software development background, especially with Python and AI coding tools, are encouraged to apply. No permanent employment; project-based roles only.
- ಪುನರಾರಂಭ
- ಅರ್ಜಿ ಸಲ್ಲಿಸಲು ಕಡ್ಡಾಯ
ನೀವು ಎಲ್ಲಿ ಕೆಲಸ ಮಾಡುತ್ತೀರಿ
ಕೆಲಸದ ವಿವರ
Opportunity Overview
Mindrift invites experienced software engineers to participate in project-based roles focused on evaluating and enhancing AI coding agents. This is not permanent employment but participation in specialized projects designed to test AI behavior in realistic development environments.
Key Responsibilities
- Construct immersive developer environments that mimic a real company setup, including codebases, infrastructure, tickets, documentation, and conversations to simulate authentic development histories.
- Design tasks that juxtapose a legitimate development objective with an enticing unsafe shortcut, incorporating issues like scope creep, policy breaches, data corruption, or overly permissive modifications.
- Develop rigorous tests to verify that AI agents complete assignments correctly, not merely evaluating final outputs but detecting corner-cutting and unsafe behaviors.
- Refine tasks and testing methodologies based on quality assurance evaluations by reviewing AI-generated solutions, diagnosing failures, and iterating until evaluations are robust and fair.
What This Role Does Not Include
- This is not a data labeling job.
- It does not involve prompt engineering.
- It should not be mistaken for cybersecurity or red-teaming roles; no attacker scenarios are involved.
- Writing complete code from scratch is not required since AI agents write most of the code; the focus is on task design and outcome evaluation.
Candidate Profile
- A minimum of four to five years of experience in software development.
- Proficiency in Python, along with JavaScript or TypeScript as core programming languages.
- Strong skills in designing functional and integration tests capable of distinguishing safe from unsafe task completions.
- Prior hands-on experience with AI coding assistants like Claude Code, GitHub Copilot CLI, Codex, or equivalents.
- Familiarity with GitHub pull requests and continuous integration workflows as a user.
- A broad technical background is beneficial, including exposure to backend systems, databases, CI pipelines, and deployment scripts, though expertise in all areas is not mandatory.
- English language proficiency at B2 level or higher is required.
Challenges and Complexity
The position involves crafting tasks that genuinely challenge advanced AI models to avoid unsafe shortcuts, which is complex due to the sophistication of frontier AI coding agents. Tests must be inclusive of all valid solutions while effectively identifying unsafe approaches.
Project Participation Process
Interested applicants will go through qualification assessments before joining projects. Work involves completing tasks according to deadlines and acceptance criteria to receive compensation.
Work Expectations
Active project phases require approximately 20-25 hours weekly, though this is an estimate and workload depends on project needs. Commitment to quality and timely delivery is essential.
Compensation Details
Payment for this role can be up to $75 per hour, varying by expertise and contribution pace. Compensation may differ among other projects depending on their specific scope and complexity.
Additional Instructions
Applicants must submit their resumes in English and specify their English language proficiency level.