LLM Inference Engineer

Make our serving stack faster and cheaper — kernels, continuous batching, speculative decoding, end to end.

• Optimize the serving path end to end
• Implement continuous batching + speculative decoding
• Own GPU utilization and cost per token

• Deep CUDA or Triton experience
• Comfortable in Rust or C++
• Track record of measurable perf wins

Veqto serves open-weight models to developers. Profitable, ~30 people, infrastructure-obsessed.