Smile

Smile (Statistical Machine Intelligence & Learning Engine), an open-source project maintained by Haifeng Li, is a comprehensive machine learning library for the JVM usable from Java, Scala, Kotlin, Clojure, or Groovy, used in production at LinkedIn, Citi, and moKredit. It covers deep learning and LLMs with GPU-accelerated neural networks via LibTorch (including pretrained EfficientNet-V2 models and a native, OpenAI-compatible LLM inference REST server), classic algorithms (decision trees, random forests, SVMs, clustering), NLP tools (tokenizers, stemmers, POS tagging, Word2Vec), mathematical utilities (matrix operations, statistical tests, symbolic differentiation), and both interactive 2D/3D plots and Vega-Lite declarative visualization. The project states it outperforms R, Python, Spark, and XGBoost on benchmarks with lower memory usage. Completely free and open source, actively maintained with version 6.3.0 released August 2026. Good fit for a JVM-based team (Java, Scala, Kotlin) that wants a genuinely comprehensive ML library — classic algorithms through GPU-accelerated deep learning and LLM inference — rather than calling out to a separate Python ML service.