Agentic AI with Quarkus, LangChain4j and vLLM
summary
Builds agentic systems with Quarkus + LangChain4j, serving models on-prem with vLLM and Kubernetes, and shows working examples: a parallel workflow that queries two LLMs (movie + food) then aggregates results, and an agent that calls three tools (city extractor, geocoding, weather) sequentially — traceable end-to-end. Runs quantized models locally (Qwen 2.5 14B and smaller llama variants) and also targets vLLM for throughput. Notes hallucinations ~10–20% failure modes, tool-invocation errors, and uses guardrails / LLM-as-judge to detect and block bad outputs.
AI-written summary · about
chapters (16)
- 0:00Introduction and motivation
- 1:05Enterprise integration
- 2:55Roles and methodologies
- 6:16AI foundations
- 7:16Transformer architecture
- 15:06Workflow patterns
- 16:51Agentic design
- 18:26Practical weather agent example
- 23:16Agentic tracing
- 24:38Search-capable agents
- 27:44Lessons learned
- 30:03Addressing hallucinations
- 32:51Jailbreaking and security
- 40:15Guardrail implementations
- 42:08Quarkus guardrail coding
- 44:58Production deployment
related talks
39mLLM Agents Gone Wild (And How to Tame Them with Quarkus)
01h 01mScaling AI with Java: Building Production-Ready LLM Applications
45mComplementing LLM with agentic AI: a Quarkus story on AI
47mQuarkus and AI Integrating Java and LLM to build intelligent applications
36mCrafting Intelligent Cloud Native applications with Quarkus
49mPushing LLMs over the Edge: Exploring the Limits of the Possible
44mCrafting Intelligent Supersonic Subatomic applications with Quarkus
01h 01mCreate #AI Infused #Java Apps with #LangChain4j