GLA Summit

We present SOTA, the first complete deep learning and computation framework built entirely on ONNX and ONNX Runtime, allowing engineers, researchers, and academic users to import, edit, train, execute, and deploy AI models and general-purpose computation graphs directly in LabVIEW, the industry-standard graphical data-flow programming language.

No Python. No Docker. No Jupyter.

Just pure data-flow programming—powered by LabVIEW, designed for those who demand modularity, determinism, and performance in their control, measurement, or AI systems.

Unlike traditional code-based environments, LabVIEW’s graphical data-flow design enables real-time interaction with variables, live updates, and intuitive debugging. This built-in interactivity offers a unique development experience where users can manipulate, observe, and annotate execution flows directly—making AI model development not just efficient, but truly interactive.

For the first time, we offer a simple graphical interface to the ONNX / ONNX Runtime ecosystem, unlocking new development approaches powered by LabVIEW’s graphical data-flow paradigm.

SOTA goes further than any existing solution.

It delivers a unified platform experience that eliminates the fragmentation typically encountered when getting started in AI development.

Not only does it support both training and inference, but it also enables intuitive construction and editing of computation graphs: users can visually build, modify, and orchestrate ONNX graphs as native LabVIEW workflows.

SOTA goes further than any existing solution:

Executing pre- and post-processing functions as native LabVIEW nodes

Running Reinforcement Learning agents with full ONNX execution graphs

Reproducing advanced deep learning workflows like YOLOv11 segmentation—entirely within ONNX Runtime, and without relying on Python or Ultralytics

This is a technological first.

By extending ONNX Runtime beyond AI—into the execution of general-purpose computation graphs—SOTA introduces a new paradigm: graph-based computing for control, test & measurement, and high-performance applications.

SOTA was built for three core domains:

Industry, where ONNX models are integrated into state machines, QMH, or Actor Frameworks for real-time automation and embedded deployment

Research, where low-level ONNX graph control enables complex architectures, custom training loops, and experimental flexibility

Academia, where the Keras-like high-level API and visual data-flow design drastically lower the entry barrier, accelerating AI education

We actively contribute to the ONNX Runtime training branch and have implemented missing operators to unlock full training support, enabling two global firsts:

Training a YOLOv11 segmentation model entirely within ONNX Runtime: https://www.youtube.com/watch?v=noDajU7TQDU

Implementing a DDPG reinforcement learning agent as ONNX graphs in LabVIEW: https://www.youtube.com/watch?v=PVkrpMrQd1s

The ONNX Runtime / LabVIEW combination empowers users to integrate AI models into any software architecture, positioning ONNX as a true runtime backend for industrial-grade systems.

SOTA was showcased at NI Connect 2025 in Fort Worth, where it received enthusiastic feedback from the LabVIEW community.

It is also available for free to students, supporting education and open research.

SOTA has been shortlisted for the EIC Accelerator, one of Europe’s most competitive deep tech funding programs—often considered the European counterpart to Y Combinator.

More on the vision behind SOTA: https://youtu.be/H796QY27Nw8?si=27EC2Skgpm3-1bR4