About the company
Our client is a software company.
The role
Raydar is recruiting for this role on behalf of our client. Build and tune the serving and training infrastructure behind specialized language models, from early experiments through to live production use. The work centers on making model execution quicker and less costly on accelerators while keeping releases dependable and observable.
What you'll do
- Lead end-to-end care of the systems that run and train custom language models, from early testing through live traffic.
- Benchmark and raise response speed, token throughput, hardware utilization and cost under production-like loads.
- Refine request grouping, caching, memory footprint and model execution, and test approaches such as reduced-precision formats and accelerated decoding.
- Keep model inference fast for evaluation runs, synthetic data generation and reinforcement learning sampling, and build dependable release and monitoring pipelines.
- Work alongside leadership and researchers to split accelerator capacity between training and serving, and to roll post-trained models out into client environments.
Requirements
What we're looking for
- Proven record building or tuning language model serving software in a live company setting, with clear credit for performance or stability gains.
- Solid low-level engineering and troubleshooting ability, including familiarity with GPU memory and compute ceilings.
- Deep familiarity with hardware chokepoints, serving tuning and observability for model workloads, including how output quality trades against speed.
- Hands-on work with a mainstream LLM serving framework or a comparable in-house engine, written in PyTorch or JAX.
- Experience owning an infrastructure system end to end, ideally built from scratch on a small team or at an AI-focused startup.
- Skill in balancing output accuracy against response time, capacity and expense when choosing pragmatic approaches.
- Willingness to move an idea from proof of concept to a live service within a compact group.
Bonus points
- Low-level kernel programming, reduced-precision numerics, accelerated decoding, or running many models across several machines.
- Background at an AI-focused company that trains its own models.
- A technical degree from a highly regarded program.
- Published research or public writing about systems work.
Benefits
Compensation and benefits
- Base salary: USD 200,000 to 350,000 per year
- Equity
Location and work model
- Bay Area, California, United States
- On-site, 5 days a week in person
- Full-time
Skills Required
- Proven experience building or tuning language model serving software in a live company setting
- Low-level engineering and troubleshooting ability, including GPU memory and compute constraints
- Deep familiarity with hardware bottlenecks, serving optimization, and observability for model workloads
- Experience with a mainstream LLM serving framework or comparable in-house engine using PyTorch or JAX
- Experience owning an infrastructure system end to end
- Ability to balance model accuracy, response time, capacity, and cost
- Ability to take ideas from proof of concept to production service
- Low-level kernel programming, reduced-precision numerics, accelerated decoding, or multi-machine model deployment experience
- Experience at an AI-focused company that trains its own models
- Technical degree from a highly regarded program
- Published research or public writing about systems work
What We Do
Raydar is a talent acquisition and business consulting firm that connects world-class and emerging talent with growing organizations. It supports companies through team development, strategic hiring, and customized growth solutions, helping clients recruit roles such as engineers, product managers, executives, legal counsel, and quantitative traders. Raydar focuses on understanding each organization’s needs, culture, and long-term goals to build high-impact teams.
