About the company: US-based AI startup focused on building the next generation of training data for LLMs. The team partners with top AI labs to create realistic RL environments where models encounter research and engineering challenges, iterate, and learn from feedback, pushing AI closer to its full potential.

Project: Design and build reinforcement learning environments to teach LLMs advanced reasoning and modern ML concepts. Candidates will work on realistic feedback loops where models encounter research and engineering problems and iterate on solutions.

• Build and maintain RL/ML environments for LLM training
• Implement robust, production-quality Python code (not just notebooks)
• Deploy and run environments in Docker with focus on reliability and iteration speed
• Analyze model performance and respond to feedback efficiently
• Collaborate with research teams to translate papers and ideas into RL problems

• Client: US startup
• Recruitment process: 2 meetings with hiring managers, followed by a phone screen with our recruiter and technical test
• Fully remote work

• Strong Python (engineering-quality)
• Docker and production mindset
• Understanding of LLMs and their limitations
• Ability to meet throughput expectations
• Advanced English (C1/C2) and ≥4 hours overlap with US time zones

• Deep knowledge of transformer internals and LLM training/inference
• Experience with inference libraries (vLLM, SGLang, etc.)
• CUDA or Pallas kernel development experience
• Publications or open-source contributions in active DL/ML research
• Experience building interactive RL environments and RL-based learning systems

• Fully remote, flexible work schedule with some overlap to US time zone
• Direct impact on how LLMs learn
• Collaboration with top AI researchers and labs
• Exposure to cutting-edge RL and ML projects