The company was a 1 person company from 0 → 10M of revenue.
Every employee should contribute >30M of revenue/yr to the company.
The best candidates would be top 1% at multiple parts of the inference stack, yet have breadth across the whole stack.
The company builds high margin inference infrastructure. Our stack spans kernels, model serving, routing, autoscaling, and capacity.
Responsibilities
Find the gap between theoretical hardware performance and production performance
Trace latency and throughput regressions from the API layer down to individual kernels
Optimize batching, scheduling, routing, quantization, and distributed execution
Work on new research directions around caching
Work with NVLink and RoCE
Validate that every optimization preserves model quality and correctness
You might be a fit if you
Have optimized complex production systems
Can juggle 8+ Codex/Claude/other coding agents concurrently
Understand GPU performance, memory bandwidth, collectives, and inference serving
Are strong in Python, CuTEdsl, and comfortable navigating unfamiliar codebases
Care about tokens per second, tokens per dollar, and correctness equally
You will work directly with the founders on problems that determine how efficiently frontier-scale models can be served. Small team, enormous compute, immediate production impact.