SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Salary: USD 400,000 - 0 / annual
Etched is building AI systems that accelerate its own development. As a Research Engineer on the RL Frontiers team, you will train proprietary frontier models to automate kernel development, ASIC engineering, production pipelines, and 3D hardware operating systems.
You will own the post-training stack completely end-to-end, building RL environments with ASIC, kernel, and supercomputing engineers. Your work will be rewarded by simulation, formal tools, tests, and hardware measurements. You will create custom data mixes that unlock frontier performance on relevant tasks and develop benchmarks that drive real productivity in models while avoiding reward hacking.
Key responsibilities include:
- Owning the post-training stack end-to-end
- Building RL environments with cross-functional engineering teams
- Creating custom data mixes for frontier performance
- Designing benchmarks that measure real productivity and prevent reward hacking
Etched is a fully in-person team in San Jose (Santana Row) with no boundaries between engineering and research. The company is building hardware for frontier intelligence, co-designing chips, racks, software, and manufacturing to deliver best-in-class throughput and latency. Backed by hundreds of millions from top-tier investors and staffed by leading engineers, Etched is redefining the infrastructure layer for AI inference.
Benefits include medical, dental, and vision packages with generous premium coverage; $500/month credit for waiving medical; $2,500/month housing subsidy for those within walking distance of the office; relocation support; wellness benefits; daily lunch and dinner; and unlimited compute budget subject to ROI justification.
REQUIREMENTS:
- Hands-on post-training experience, including SFT and RL (e.g., PPO, GRPO), with shipped models that people used
- Demonstrated ability to take a model from poor to good performance on tasks it wasn't built for, ideally with scarce data (e.g., as an early post-training hire at an applied AI company)
- Strong PyTorch and distributed-training engineering skills; ability to read rollouts and verify numbers before trusting results
- Understanding of the systems side of RL to make the loop faster and cheaper, not just better
- Entrepreneurial mindset and good decision-making under compute and data constraints (former founders are a plus)
STRONG ADDITIONAL EXPERIENCE:
- RL or post-training frameworks (verl, OpenRLHF, TRL, prime-rl) or inference engines (vLLM, SGLang)
- Agentic, multi-turn RL with verifiable rewards
- Distillation and training small, specialized models for production
- ASIC design or verification, accelerator kernels, or large-scale compute infrastructure
