Skip to content

Conclusion

Alright, this is the end! We hope you enjoyed this workshop and gained valuable insights into building AI-infused applications and agentic systems.

Throughout the workshop, we followed the Miles of Smiles car rental company across three sections, building progressively more sophisticated AI-powered systems with Quarkus and Quarkus LangChain4j. Here’s a recap of what we covered:

Section 1 - AI-Infused Applications

  • Integrating a large language model (LLM) seamlessly within a Quarkus application
  • Utilizing annotations to efficiently pass prompts and structure interactions
  • Implementing the Retrieval Augmented Generation (RAG) pattern to enrich responses with external data
  • Leveraging function calling to create tools that LLMs can reason over and invoke
  • Integrating remote tools and services via the Model Context Protocol (MCP)
  • Implementing guardrails to safeguard against common risks, such as prompt injection and LLM misbehavior
  • Adding observability and fault tolerance

Section 2 - Agentic Systems

  • Integrating AI agents into a Quarkus application in a similar way to AI services
  • Connecting agents into chains using sequence workflows with shared state
  • Invoking agents in parallel workflows to perform work more efficiently
  • Building conditional workflows that let you control which agents work on a request
  • Combining agents and workflows of agents into nested workflows
  • Keeping a human in the loop for approval and intervention
  • Processing multimodal inputs such as images alongside text
  • Engaging remote agents, potentially built using different agentic frameworks, using Agent-to-Agent (A2A) communication
  • Deploying a multi-agent system to Kubernetes/OpenShift

Section 3 - Enterprise Agentic AI Patterns (experimental)

  • Defining and dynamically discovering agent skills at runtime
  • Enforcing guardrails and compliance policies across agents
  • Building event-driven workflows with Quarkus Flow
  • Persisting workflow state across restarts with PostgreSQL
  • Applying voting and loop patterns with adaptive model selection
  • Composing custom orchestration with a PlannerAgent
  • Integrating external tools via MCP
  • Communicating with remote agents via A2A
  • Evaluating agent behavior with LLM-based testing

By the end of this workshop, you should have a solid foundation for building AI-enhanced applications with Quarkus, from a simple chatbot all the way to an enterprise-grade agentic system. If you have any questions or feedback, don’t hesitate to reach out to us on Zulip or open a discussion on GitHub. We’re excited to see what you build next!