How ZoomInfo Identified and Eliminated 4am OOM Crashes with AI

The team naturaly wanted higher stability. What made it difficult wasn’t the existence of the ELU spike. It was the lack of visibility into what was causing it. The spike occurred consistently, but existing tools gave sparse, disparate data that required tons of work by the senior engineers on the team to figure out.

ZoomInfo didn’t need another metric. They needed the root cause. Adding Hud to the service reviewed exactly what was going on, using function level visibility and forensic data straight from that 4am slowdown. The team was quickly able to figure out the issue was massive invocations of a dependency lookup, and quickly fix it with Claude Code to achieve: