Java Vienna

save the date!

Agenda:
18:00 Doors open
18:20 welcome - doors closing
18:30 "living immutable: working with Java Records the painless way" - Georgios Angelopoulos
19:30 break/networking
20:00 Zero-Dependency Java 25+ AI Business Agents: Running locally with zsmith and LightMetal - Adam Bien

living immutable: working with Java Records the painless way
Java Records are great for immutability, but updating deeply nested structures usually means dealing with a headache of nested constructors and endless .withX() chains.

This talk covers how to skip the boilerplate using Telescope, a lightweight Java library designed for painless deep updates. We will dive into:

• The ergonomic pain points of nested Java Records.
• Using Telescope's fluent DSL for surgical, zero-boilerplate updates
• Applying these same clean immutability patterns to legacy POJOs.

About Georgios Angelopoulos
I am a Lead Developer at Prewave with a background in Electrical and Computer Engineering. My experience and interests cover the entire computing spectrum, solving problems at every layer of the stack.

Zero-Dependency Java 25+ Business Agents: Running locally with zsmith and LightMetal
Modern Java provides all the functionality required for local LLM inference and agent orchestration, eliminating the need for external dependencies.

LightMetal directly binds Llama.cpp from Java using the Foreign Function and Memory API (FFM), eliminating the need for JNI or native wrappers. It runs open models such as Mistral, Gemma, and Llama on your machine.
Zsmith is built on LightMetal and uses the Boundary-Control-Entity (BCE) model to structure agents. It loads tools and skills as ServiceLoader plugins and runs tool calls and streaming on Virtual Threads. Local inference and remote models work with the same agent code over the JDK's HttpClient.

Topics:
- One-line installation, no build tools, no package managers
- FFM in practice: calling native libraries without JNI
- Running Mistral, Gemma and Llama locally in pure Java
- ServiceLoader as a plugin mechanism for tools and skills
- BCE applied to AI agents
- Virtual Threads for streaming, tool calls and parallel inference
- Switching between local and remote models without changing the agent
- Plain JDK: HttpClient, records, sealed types and pattern matching
- Packaging agents as single-file Java applications

Everything is demonstrated live with code. Your questions are highly welcome — the more, the better!