AI Incident Triage Agent

A Spring Boot portfolio project for testing local AI incident triage on fake tickets, logs, health snapshots, and runbooks.

3 fake production incidents

3/3 expected causes caught

0 real tickets or private logs

Why This Project Exists

An AI incident helper should not guess. It should read evidence, compare it to a runbook, explain the likely cause, and suggest a safe next action. This project tests that behavior with fake incidents.

Incident Cases

Case	Service	Expected cause
`INC-001`	`checkout-api`	Database connection pool exhaustion.
`INC-002`	`auth-service`	Redis login cache node is unreachable.
`INC-003`	`media-worker`	Temporary upload volume is full.

First Local Run

Case	Score	Grade
`INC-001`	1.0	pass
`INC-002`	1.0	pass
`INC-003`	1.0	pass

The local model caught all three expected causes and gave useful evidence-backed next actions.

Local Workflow

./mvnw spring-boot:run
scripts/run_triage.sh INC-001
scripts/score_triage.sh INC-001 results/INC-001-triage.md

Portfolio Story

Built a Spring Boot incident-triage evaluation app using fake production tickets, logs, health snapshots, runbooks, read-only tool endpoints, local LLM prompts, REST scoring endpoints, and a Hugging Face-ready static report.

Next Step

Wrap the read-only tool endpoints with a real MCP server and compare two local models on the same incident set.