LLM Agents Gone Wild (And How to Tame Them with Quarkus)
summary
Demo shows a Quarkus + LangChain4j agentic module that generates implementations at build time and wires agents by annotations, uses an agentic scope map keyed by required output key, and supports workflows (sequence/parallel/loop) or pure supervisor agents. Observability uses an agent monitor HTML report and OpenTelemetry; guardrails enforce input/output guardrails with retry/reprompt; fault tolerance uses MicroProfile annotations; testing uses an evaluation framework and judge models. Demo app (Dr. Banner vs Hulk) runs but the supervisor sometimes skips verification and type parsing from LLMs intermittently fails.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:45LLM agent risks
- 4:17LangChain4j agents in Quarkus
- 7:52Agent workflows
- 13:16Agent patterns
- 16:29Demo: Hulk vs Dr. Banner
- 19:43Enterprise agent patterns
- 24:03Observability with guardrails
- 27:36Fault tolerance testing
- 32:11Demo: Supervisor & human-in-loop
- 36:33Summary
related talks
48mAgentic AI with Quarkus, LangChain4j and vLLM
49mPushing LLMs over the Edge: Exploring the Limits of the Possible
45mComplementing LLM with agentic AI: a Quarkus story on AI
53mHigh Speed Java Development with LLMs and Agents Fast, Clean, Maintainable
52mLivecoding: Creating Beautiful Java Code With LLM and Agents
49mAgentic AI Patterns
36mCrafting Intelligent Cloud Native applications with Quarkus
44mUnlocking the Future of Java: LLM Integration Unveiled by Quarkus & LangChain4j