FlexibleSan Francisco
Research Engineer
О роли
Описание вакансии
About the Team
- OpenAI develops models that can reason through complex problems and hardware designed for advanced AI.
- AI for Chips applies increasingly capable AI systems to semiconductor engineering.
- Goal is to help engineers develop better chips and shorten design cycles.
- Brings research, model training, and hardware expertise together.
About the Role
- Hiring a Research Engineer to help OpenAI models solve chip-design problems through reinforcement learning, tool use, and evaluation.
- Own experiments from initial idea through implementation and analysis.
- Build environments and evaluations, run training, investigate failures, and use results to decide next steps.
- Build software needed to make experiments reliable and reproducible.
- Prior chip-design experience helpful but can learn domain alongside hardware specialists.
Responsibilities
- Build RL environments and evaluations for tasks such as RTL generation, design verification, and physical design optimization.
- Develop and test approaches that help models use chip-design tools and improve power, performance, and area while preserving correctness.
- Design experiments, establish baselines, and measure improvements on new tasks and designs.
- Investigate failures across model behavior, rewards, evaluation tools, and experiment infrastructure.
- Improve iteration speed through better tooling, faster evaluations, and proxy rewards.
- Turn successful experiments into reusable research code and training workflows.
Requirements
- Strong programming and debugging skills and a track record of turning technical ideas into working software.
- Experience with reinforcement learning, model evaluations, post-training, or other applied ML research.
- Experience building tool-using agents, reward functions, or automated evaluation systems.
- Can form clear hypotheses, design useful experiments, and distinguish meaningful results from noise.
- Work independently on ambiguous problems and make practical decisions.
- Stay close to the implementation and can explain what you built, what failed, and what you learned.
- Communicate progress clearly and collaborate well with people across research, software, and hardware.
- Care about developing safe, beneficial AI.
Nice to have
- Familiarity with experiment orchestration, distributed training, or research infrastructure.
- Experience with RTL, Verilog/SystemVerilog, EDA tools, formal verification, or chip-design automation.
Стек и навыки
С чем работаем