Member of Technical Staff, Inference

About Hopper

Hopper serves open-weight speech and language models for voice applications. We build the infrastructure behind them, from model runtimes and GPU scheduling to streaming APIs.

About the role

We’re hiring an inference engineer to improve latency, throughput, and reliability across our serving stack. You’ll profile workloads, improve serving code, and take changes through to production.

This is a full-time, in-person role. You’ll work from our beautiful office in San Francisco’s Financial District.

Responsibilities

• Optimize batching, caching, and scheduling for speech and language models.

• Profile GPU workloads and resolve bottlenecks in model execution and streaming.

• Build benchmarks for latency, throughput, and model quality.

• Deploy serving improvements and troubleshoot production issues with customers.

Qualifications

• Experience building and operating inference or distributed systems.

• Ability to trace performance problems across the stack and validate improvements with measurements.

• Experience with PyTorch, vLLM, or SGLang. CUDA or Triton experience is a plus.

Compensation and benefits

Annual base salary: $300,000–$350,000 USD, plus equity.

• Daily lunch and dinner in the office.

• Unlimited office book budget.

• Unlimited compute budget, subject to ROI justification.

• $1,500/month housing stipend for living within walking distance of the office.

• Relocation and visa sponsorship support.

• Free Equinox membership.