Senior Machine Learning Engineer, Jockey Core

Seoul, South Korea

About the role

Build benchmarks and load tests replaying agent real traffic, measuring TTFT and inter-token latency separately. Apply inference optimization techniques (quantization, batching, disaggregated prefill/decode, speculative decoding) to hit cost and latency targets. Build cost models and drive production hardening through autoscaling, capacity planning, observability, failover, and rollback-safe rollouts.

Requirements

• Significant experience serving and optimizing large-scale LLM inference in production (vLLM, TensorRT-LLM, SGLang, or similar).
• Experience with batching, scheduling, quantization, disaggregated prefill/decode, and speculative decoding techniques.
• Experience designing and operating large-scale distributed systems in high-performance GPU environments.

Nice to have

• Experience contributing to or customizing internals of LLM inference servers (vLLM, TensorRT-LLM, SGLang, or similar).
• Understanding of serving characteristics of compressed (pruned or quantized) models or reasoning or agentic LLMs.
• Experience designing multi-region or multi-cluster serving infrastructure or large-scale GPU capacity planning.

Context

Foreign-invested AI company (San Francisco HQ, Seoul R&D office) with strong visa sponsorship track record; role is fully English-language, part of global Cognition Models team, likely F-series visa pathway (F-2 / F-4 / F-5 / F-6 holders welcome). Korean language not required.