Morph, a YC‑S23 startup in San Francisco, hires a performance engineer to accelerate its inference infrastructure. Candidates should rank in the top 1% across the stack, from kernels to autoscaling. The role offers $175K - $350K and direct collaboration with founders to push frontier‑scale models.

Responsibilities include identifying the gap between theoretical hardware performance and production, tracing latency and throughput regressions from the API to individual kernels, and optimizing batching, scheduling, routing, quantization, and distributed execution. Engineers build benchmarks and observability tools to expose bottlenecks, validate that optimizations preserve model quality, and ship high‑impact fixes.

Ideal candidates have experience optimizing complex production systems, deep knowledge of GPU performance, memory bandwidth, collectives, and inference serving, and strong Python skills. They should translate profiling data into actionable decisions, balancing tokens per second, tokens per dollar, and correctness. The small team works on problems that directly impact how efficiently large models are served.

Morph develops specialized code‑generation models, serving them on a custom stack that includes autoresearch for kernels and speculative‑decoding models. Founded in 2025, the company remains active with a core team of three.