SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Sail Research is building the world's most efficient software for LLM token inference and agent hosting. As a Member of Technical Staff - Inference, you will own token processing across the full stack, from high-level optimization strategies down to GPU kernel implementation.
You will modify and extend state-of-the-art inference engines like vLLM and SGLang, optimizing every microsecond of GPU time during forward passes. You'll design and implement novel parallelism schemes tailored to diverse hardware topologies, and write custom GPU kernels for specialized regimes such as cascade attention. Your work will involve deep profiling with tools like NSys, understanding request scheduling at the engine level, and maximizing efficiency across heterogeneous hardware.
The ideal candidate has a strong understanding of core LLM mechanics including KV cache, mixture-of-experts, and prefill vs. decode phases. You should be familiar with modern tile-based GPU programming frameworks such as Triton, CUTLASS, or ThunderKittens, or eager to learn them. You'll follow MLSys research closely, implementing cutting-edge ideas like speculative decoding and sparse attention. Excellent interpersonal communication is essential—the team values clear, human-written prose and collaboration.
The interview process includes technical discussions with the CTO and CEO, an in-person interview day at Sail's San Francisco office, and a 3-4 hour hands-on coding challenge that simulates real production work. You'll receive immediate feedback on your solution and present your process and learnings to the team. AI tools are encouraged during the technical assessment.
Sail operates from a modern downtown San Francisco office with comprehensive perks including catered meals, Studio Displays at every desk, and a friendly office cat named Coco.
