Daniel Kim is Head of Growth at Cerebras Systems. Previously he led developer relations at New Relic, and his public work explains how inference infrastructure affects the speed of AI applications.[1]
Start with How Inference Keeps Up with Our Insatiable Need for Tokens — Cerebras & Gimlet Labs →
Links
Personal website
X profile 
GitHub · lazyplatypus Head of Growth · Cerebras
Career
Kim’s personal site identifies his current Cerebras role and previous leadership of developer relations at New Relic. It links his public GitHub and social accounts. These roles place his technical explanations in the context of helping developers understand and adopt infrastructure products.[1]
Contributions
In a joint session with Gimlet Labs cofounder Michelle Nguyen, Kim explains why reading a prompt and generating tokens put different demands on hardware. Batching fills spare computation with other requests; cache techniques manage and reuse request state; separating prompt processing from generation lets each phase use a suitable worker pool. The talk connects these scheduling choices with the repeated model calls made by coding agents, where waiting on each call accumulates.[3]
Tools, writing & talks
Writing, research & talks
Start here
A joint explanation of batching, cache management and matching inference phases to hardware.
Start with the distinction between prompt processing and token generation, then follow the scheduling examples. Basic familiarity with language-model serving helps; vendor performance figures describe the systems and configurations discussed in the talk.
Copresenter with Michelle Nguyen[3]Original source