About this talk
This talk, presented by Luqman Saeed from Payara, focuses on advancements in embedding AI directly into Jakarta EE applications through a local-first architecture, showcased at Open Community Experience 2026 in Brussels. It explores how to integrate AI capabilities without relying on external APIs, thus maintaining control over data and costs. The implementation leverages a Java-centric stack, utilizing CDI for dependency injection, LangChain4j for AI orchestration, and PostgreSQL with pgvector for embeddings.
The session delves into four architectural patterns, including retrieval-augmented generation (RAG) pipelines and agent-based workflows, which enhance flexibility while minimizing dependency on external services. Key trade-offs concerning model size, performance, and operational complexity are discussed, particularly differentiating between external inference with Ollama and in-JVM execution using Jlama. This approach addresses critical practical concerns such as latency, vendor lock-in, and regulatory constraints, ultimately highlighting the benefits of seamless AI integration within enterprise applications.
