SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Normal Computing is an applied AI company building foundational hardware and software for the semiconductor industry and critical AI infrastructure. This Research Engineer role focuses on advancing agentic LLMs and reinforcement learning for Normal EDA, an agentic AI platform for semiconductor design automation.
You will design and execute experiments to build multi-agent systems for code generation that interact with EDA tools including simulations, waveform analysis, formal verification tools, and physical design tools. The role involves proposing fixes and iterating through all stages of chip design and verification flows. You'll create research prototypes that integrate with the production agentic code generation tool and collaborate with engineering to productionize successful research outcomes.
Key responsibilities include creating RL environments and evaluations for agents, exploring proxy rewards, and balancing speed/accuracy tradeoffs of custom tools. You'll generate datasets from silicon collateral such as RTL designs, testbenches, custom VIPs, chip specifications, and agent logs, with synthetic data generation where appropriate. You'll maintain data cards and manage licensing considerations.
The role requires disciplined experimental analysis with rigorous ablations, clear documentation of results, and staying current on LLM agents, reinforcement learning (offline/online, RLHF/RLAIF), constrained decoding, and program synthesis.
Ideal candidates have a PhD in CS/AI/ML or equivalent research experience with publications in multi-agent RL, agentic AI, or RL for language/code. Strong Python and ML framework experience (PyTorch preferred) is essential, along with demonstrated ability to turn research into working systems. Experience designing evaluation environments and reward models for sequential/agentic tasks is required. Fluency with EDA tools (formal, simulation, physical design) and comfort with data acquisition/curation are important. Bonus qualifications include research in program synthesis/codegen, constrained decoding, execution-based rewards, offline RL from tool traces, open-source contributions, semiconductor domain familiarity, and a track record of shipping research to production.
