AI vision software company Neurolcle announced that it unveiled its deep learning training engine, ‘Neuro-T Engine,’ for the first time at the “Neurolcle AI Machine Vision Forum” held during the AW2026 exhibition.

The newly introduced Neuro-T Engine is a field-oriented AI training engine designed to enable on-site training and operation of AI models directly within manufacturing environments.

In recent manufacturing industries, as processes become more advanced and products more sophisticated, defect patterns are becoming increasingly fine and irregular. In addition, frequent occurrences of new defect types and changes in process conditions have led to growing demand for continuous updates and retraining of AI vision inspection models.

However, conventional AI vision systems often provide model training functions as separate development environments or standalone software, making it difficult to perform training directly within production equipment or systems. As a result, data acquired on-site had to be transferred to external servers or development environments for retraining and then redeployed, leading to slower response times and increased operational burden.

▲ Image of the deep learning training engine ‘Neuro-T Engine’

The newly released Neuro-T Engine was developed to address these structural limitations and is provided in the form of a CLI (Command Line Interface) and API (Application Programming Interface). By embedding only the training module into existing inspection systems or equipment, it can be easily applied, enabling the addition of AI model training capabilities without requiring changes to the existing environment.

Through this, machine builders (MB) and system integrators (SI) can efficiently integrate AI training capabilities without modifying existing product architectures. In addition, the high-performance Auto Deep Learning algorithms validated in Neurolcle’s existing products can be directly utilized, enabling high-performance AI model training without the need for separate algorithm development or optimization.

Neurolcle stated that with the introduction of Neuro-T Engine, it has completed its software lineup covering the entire lifecycle of AI vision models—training, inference, and operation—including the GUI-based model training software ‘Neuro-T,’ the runtime library ‘Neuro-R’ for executing trained models, and the on-site training engine ‘Neuro-T Engine.’ All software components are provided as modular units, allowing rapid deployment without additional development and flexible application across various manufacturing environments.

In addition, Neurolcle presented an MLOps framework optimized for manufacturing environments based on Neuro-T Engine. With its modular architecture, the solution can be deployed with minimal changes to existing systems while offering high flexibility in system architecture design.

For example, it supports configurations where AI inference and model retraining are performed simultaneously within production equipment, as well as architectures where inference is executed on the equipment while images collected from production lines are transmitted to a central server for model training and management. In this way, it can be applied to both decentralized environments where models are operated individually per equipment and centralized environments where models across multiple production lines are managed collectively.

Furthermore, Neuro-T Engine provides various training configuration options to maximize GPU resource utilization. Multiple GPUs can be used in parallel for a single model to accelerate training on high-resolution images, or GPU resources can be partitioned to train multiple models simultaneously. It also supports virtual GPU-like functionality, enabling efficient resource utilization even in constrained hardware environments.

A participant at the event commented, “To stably operate AI models in manufacturing environments, an MLOps framework that supports not only inference but also rapid training and updates is essential,” adding, “Neuro-T Engine allows AI training capabilities to be easily embedded while maintaining existing system architectures, making it highly practical for building a field-oriented AI training infrastructure.”

Hongseok Lee, CEO of Neurolcle, stated, “With the launch of Neuro-T Engine, we expect to enable more efficient deployment of AI vision MLOps environments optimized for industrial sites.” Further details about Neuro-T Engine can be obtained by contacting Neurolcle.