incident-triage-agent

incident-triage-agent is a tool-calling AI agent lab built with Spring AI. It is a lean, focused Spring Boot repository demonstrating tool-calling AI agents and structured outputs using Spring AI. The project evaluates AI agent patterns at the JVM level, including how an LLM decides which Java methods to call, how Spring AI orchestrates that loop, and how the final answer is deserialized into a typed Java object. It deliberately avoids over-engineering and does not claim production readiness.

The repository provides one real, end-to-end tool-calling agent built entirely on Spring AI's own abstractions. It shows how to expose plain Java methods as LLM-callable @Tools with Spring AI, how ChatClient drives the request → tool-call → tool-result → response loop, and how to get strictly-typed structured output (a Java record) back from a non-deterministic LLM response instead of parsing free text. It also demonstrates how retrieval-augmented generation (RAG), even a minimal in-memory one, fits alongside simpler keyword search, and when an agent might prefer one over the other.

The project emphasizes keeping an LLM honest about its own work: the reported evidence is recorded from the tool calls that actually ran, not copied from the model's narrative. It also aims to keep an AI-integrated Spring Boot service testable with fast unit tests with no network calls, plus an opt-in real-model integration test, and hooked up to a real CI pipeline covering formatting, static analysis, coverage, and license/vulnerability policy.