Writing

Notes from platform work

Practical details from backend systems, delivery, and operating software.

2026-07-29

Bound vLLM request work before it reaches the engine

Define route, byte, fanout, transformation, and tenant work limits that reject oversized vLLM requests before backend allocation.

  • vllm
  • llm-inference
  • application-security
  • admission-control
  • platform-engineering