Member of Technical Staff, Reinforcement Learning Infra

About the role

We're looking for engineers and scientists to design, optimize, and maintain the core systems that enable scalable, efficient reinforcement learning for large models. This role sits at the intersection of research and large-scale systems engineering: you'll wear many hats, from optimizing rollout and reward pipelines to enhancing reliability, observability, and orchestration, collaborating closely with researchers to make RL stable, fast, and production-ready.

Responsibilities

• Design, build, and optimize the infrastructure that powers large-scale reinforcement learning and post-training workloads.
• Improve the reliability and scalability of RL training pipelines, distributed RL workloads, and training throughput.
• Develop shared monitoring and observability tools to ensure high uptime, debuggability, and reproducibility for RL systems.

Requirements

• BS/MS/PhD in Computer Science, Engineering, or a related field (or equivalent experience).
• Understanding of ML frameworks (PyTorch, TensorFlow, Ray, Megatron) from a systems perspective.
• Experience working with reinforcement learning workloads (PPO, DPO, RLHF, or reward modeling).
• Experience with containerization (Docker), orchestration (Kubernetes), and CI/CD pipelines.

Skills

• Experience building and maintaining large-scale language models with tens of billions of parameters or more.
• Experience with ML workflow orchestration tools (Kubeflow, Airflow).
• Background in performance optimization and profiling of ML systems.

Pay

The annual base salary range for this role is $200,000 – $350,000 USD. Final compensation is determined based on experience, skills, and qualifications. Equity and benefits are included in the total package.

Benefits

• Flexible vacation and paid time off (PTO)
• Health, dental, and vision insurance
• 401k match
• Catered meals (breakfast, lunch, & dinner)
• Commuter subsidies

About Us

Inception creates the world’s fastest, most efficient AI models. Today’s autoregressive LLMs generate tokens sequentially, which makes them painfully slow and expensive. Inception’s diffusion-based LLMs (dLLMs) generate answers in parallel. They are 5x faster and more efficient, while delivering best-in-class quality. Inception was co-founded by Stanford professor Stefano Ermon, who co-invented such breakthrough AI technologies as diffusion models, flash attention, and DPO, UCLA professor Aditya Grover, who co-invented node2vec, decision transformers, and d1 reasoning, and Cornell professor and Afresh co-founder Volodymyr Kuleshov, who co-invented MDLM and Block Diffusion. We pioneered the application of diffusion to language, with world’s first (and only) commercially available dLLM, Mercury. We are currently deploying our large-scale diffusion LLMs at Fortune 500 companies. Diffusion is the technology behind today’s image and video AI, and we’re making it the standard for LLMs as well.

Our team includes engineers from AWS, Google DeepMind, Meta AI, Microsoft, HashiCorp, and OpenAI. Based in Palo Alto, CA, we are backed by top-tier venture capitalists, including Menlo Ventures, Mayfield, M12 (Microsoft’s venture fund), Snowflake Ventures, Databricks, and Innovation Endeavors, and by tech luminaries such as Andrew Ng, Andrej Karpathy, and Eric Schmidt.