Design and implement a trace collection system for distributed LLM workloads
Validate that collected traces accurately reflect real workload behavior
Integrate with and instrument major LLM frameworks to extract meaningful execution data
Use collected traces as input to discrete event simulations
Analyze trace data to surface bottlenecks and inefficiencies across the stack

3+ years of experience in AI systems, ML infrastructure, or a closely related area
Hands-on experience with at least one major LLM serving or training framework
Strong proficiency in Python and C++
Solid understanding of GPU architecture, memory bandwidth, and the difference between compute-bound and memory-bound operations
Solid understanding of distributed communication
Familiarity with parallelism strategies and how they shape execution behavior across large clusters
Open source contributions or published research in relevant areas will definitely be appreciated
Previous startup experience is a plus

AI Trace Generation Engineer

Key skills