Everpure / June 2026 – Sept 2026

AI Infrastructure Triage Agent

Built infrastructure tooling and a custom agent harness for an internal AI triage agent, reducing false positives by 89% and increasing accuracy by 45%.

resume
timelineJun 2026 - Sep 2026
focusai, devtools, infrastructure
stack
  • Python
  • TypeScript
  • MCP
  • LoRA

problem

The internal AI triage agent needed to distinguish infrastructure failures from FlashBlade simulator code issues more accurately while avoiding the cost and extensibility limits of its existing agent SDK.

approach

Built an infrastructure metrics MCP server and a custom agent harness that converts MCP servers into first-party tools and supports LoRA-adapted and fine-tuned open-source models.

implementation notes

  • Built an infrastructure metrics MCP server for an internal AI triage agent to more accurately differentiate infrastructure issues from FlashBlade simulator code issues.
  • Reduced false positives by 89% and increased the triage agent's accuracy by 45%.
  • Designed and built a custom agent harness to replace the Claude Agent SDK, reducing token costs and enabling a modular, extensible toolset.
  • Converted MCP servers into first-party tools and added support for LoRA-adapted and fine-tuned open-source models.

impact

  • Reduced false positives by 89%
  • Increased triage accuracy by 45%
  • Reduced token costs with a custom agent harness
  • Enabled a modular toolset and support for adapted open-source models
false positive reduction
89%
accuracy increase
45%