AI runs where your business runs – in Java.
Deep Netts builds the native AI execution layer for the JVM – founded by the team behind Neuroph and the JSR 381 Java standard.
Demonstrated by Oracle on the JavaOne 2026 keynote stage
Watch the keynoteIn production at a US Department of Energy lab – 1000× faster data analysis
Jefferson Lab caseThe reference implementation of JSR 381, Java's machine learning standard
JSR 381 at jcp.orgGPU-accelerated deep learning on the JVM – no Python, no native sidecars
See the platform
Deep Netts is a next-generation AI platform built entirely in Java, designed to bring advanced deep learning directly into the JVM – where most enterprise systems already run.
AI runs where your business runs – in Java.
We help organizations train, optimize, and deploy AI models without the complexity of bridging to Python or other ecosystems. With GPU acceleration, Vector API integration, and enterprise-grade observability, Deep Netts delivers the performance, scalability, and control that modern businesses demand.
By combining years of research and engineering expertise, Deep Netts enables companies to unlock the full potential of AI with faster training, lower cloud costs, and improved energy efficiency. From scientific research and manufacturing to retail and finance, enterprises use Deep Netts to power intelligent systems that are efficient, secure, and production-ready.
We are proud to collaborate with leading organizations including the U.S. Department of Energy's Jefferson Lab, University of Minnesota, and the Java Community Process, helping advance the next era of Java-native AI.
At Deep Netts, we believe that AI should fit the enterprise – not the other way around.
Collaborating with
- U.S. Department of Energy – Jefferson Lab
- University of Minnesota
- Java Community Process
From a university lab to the Java standard for AI.
The idea for Deep Netts was born in 2022, while working on Neuroph, an education software for neural networks. After achieving strong results on this project with his university team, Deep Netts' co-founder Zoran Ševarac was inspired to seek a similar solution for the enterprise environment.
Recognizing the challenges enterprises face in integrating AI into their Java systems, Zoran envisioned a platform that facilitates seamless AI integration, transforming high-fixed-cost solutions into scalable operational expenses.
Zoran Ševarac, a Java Champion, esteemed professor at the University of Belgrade, and AI researcher, is a Duke's Choice Award winner with over 20 years of experience in AI and software development, and co-author and expert group co-lead of the Visual Recognition API (JSR381) standard for machine learning-based image recognition that determines how Java applications talk to AI.
With his passion and deep understanding of neural networks, deep learning, and software engineering, combined with his extensive experience leading teams of experts, Zoran has managed to bridge the gap between AI complexities and the needs of Java developers across various sectors, including business, academia, and government. As presented at the JavaOne 2025 Conference in San Francisco, Deep Netts has brought a breath of fresh air to the Java community, empowering modern Java for the AI revolution.
But Deep Netts isn't just about technology; it's a purpose-driven company committed to accelerate implementation of AI in the world by empowering major industries and sectors in the AI race, while advancing democratization and positive societal impact of AI.
2004
Doctoral research on a Java neural network framework – the seed of Neuroph.
2013
Neuroph wins the Duke's Choice Award at JavaOne; Zoran Ševarac named Java Champion.
2017
JSR 381 Visual Recognition API launched at the JCP, co-led with Frank Greco.
2022
Deep Netts Inc founded. JSR 381 final release, with Deep Netts as reference implementation.
2023
JCP Member of the Year award for the work on Java and machine learning standards.
2025
JavaOne San Francisco: 5000× faster scientific data analysis at Jefferson Lab.
2026
Featured in the JavaOne keynote. Deep Netts 4.0 brings FFM and Vector API acceleration.
Built by people who shaped Java's AI story.
Two Java Champions, a seat on the JCP Executive Committee and the standard API for machine learning on the JVM – backed by two decades of research, teaching and enterprise practice.
Work with Deep Netts
Start a Project
Bring AI into your Java systems with our team alongside yours – from first model to production deployment.
Talk to UsTry It Yourself
Download the Deep Netts library and build your first model today. Free for small production use.
Free DownloadSee What's Possible
Explore how teams use Deep Netts in production – from fraud detection to scientific computing.
Explore Use CasesWhere to find us.
Research and engineering are run from Belgrade, Serbia, alongside the University of Belgrade AI Lab. Deep Netts Inc is registered in Delaware.
- Deep Netts Inc
- 16192 Coastal HighwayLewes, Delaware 19958, USA
- Computer programming (NAICS 6201)
- Deep Netts DOO
- Jevremova 4111158 Belgrade, Serbia
- Engineering & research
Bring high-performance AI into your Java architecture.
Evaluate Deep Netts on your workload. See what Java-native AI can do.