Writing GPU-Ready AI Models in Pure Java with Babylon

Explore how to build and run AI models directly in Java using Project Babylon in this comprehensive conference talk from Devoxx Belgium 2025. Learn to eliminate dependencies on Python or external model files by defining machine learning logic in plain Java code through Code Reflection capabilities. Discover how to leverage the Foreign Function and Memory (FFM) API to connect Java applications to native runtimes like ONNX for GPU-accelerated inference, enabling high-performance execution of LLMs, image classifiers, and object detection algorithms. Master the Heterogeneous Accelerator Toolkit (HAT) for writing and composing compute kernels in Java, empowering Java libraries to harness GPU power for intensive computational tasks. Gain practical insights into integrating AI capabilities seamlessly into the Java ecosystem, with demonstrations covering facial emotion recognition, ONNX runtime integration, and GPU workload execution. Understand the architectural advantages of running AI workloads on GPUs and explore real-world implementation examples that bridge the gap between traditional Java development and modern AI requirements.