Providing an easy-to-use environment for even beginners to develop on-device AI.
Web-based sensor data collection, preprocessing, model creation, and training without code.
Beyond a simple lightweight framework, AI can be applied to actual products.
[Editor's Note] AI is evolving into a technology that makes real-time decisions on embedded devices, rather than in the cloud. Meanwhile, on-device AI faces a high barrier to entry due to its complex development environment, and model lightweighting and debugging in MCU environments are particularly challenging. MDS Tech, therefore, provides an end-to-end development flow, enabling even non-experts to implement the technology in real-world applications, from model design to training, optimization, MCU integration, and debugging. Accordingly, we met with Taejun Park, MDS Tech's representative who will be presenting at the '
2025 e4ds Tech Day ' on September 9th, and heard about on-device AI development.
■ The biggest burden when developing existing on-device AI is The biggest burden in on-device AI development is the high barrier to entry resulting from the complexity of building a development environment.
Because the process of training AI models, optimizing them for embedded devices, and then integrating them back into firmware is performed across different tools, languages, and platforms, model lightweighting and debugging in MCU environments with limited hardware resources requires a lot of trial and error.
In particular, from the developer's perspective, it was a structure that consumed excessive time and money.
■ I am curious about the core value of MDS Tech's on-device AI platform. Also, what are the differences compared to the lightweight AI frameworks or compilers already on the market?
MDS Tech's on-device AI development solution provides a development environment focused on easy integration and practical application.
Beyond a simple lightweight framework, the entire cycle from model design → training → optimization → MCU code integration → debugging has been structured into a single flow, increasing the feasibility of applying AI to actual products even for non-experts.
While other solutions focus on model optimization and runtime, our end-to-end workflow, which integrates various third-party solutions, enables firmware developers to confidently leverage AI in their environment, which is a key differentiator.
■ You said that it helps even non-experts or AI beginners to easily design models. What functions does it have? There are a variety of third-party tools that provide sensor data collection, preprocessing, model creation, and training via a web-based GUI without any code implementation, and Keil can integrate with these and export them directly to your firmware project.
Additionally, you can get started without prior knowledge of AI model design through pre-trained blocks (Transfer Learning), AutoML, and signal processing block recommendations.
Additionally, MDS Tech provides Keil project samples, dedicated device setup, and debugging guides to minimize trial and error for beginners.
■ Please introduce the method of introducing MDS Tech solutions and examples of cooperation with customers. Introduction is carried out in two main ways.
First, the samples provided with the purchase of the package solution. Customers can apply it autonomously based on examples and technical guides.
Second, there is also a method of conducting PoC together with MDS Tech and setting up the project environment together.
■ What topics will be presented at the 2025 e4ds Tech Day? This presentation will demonstrate the entire process from developing an AI model using third-party machine learning tools and Keil to integrating it into an MCU and running it.
In particular, we will highlight how complex AI development can fall into the realm of firmware developers, focusing on practical, applicable examples.
■ A word to readers AI is no longer confined to grand cloud technology; it has now become a technology that makes real-time decisions on embedded devices.
MDS Tech is ready to help developers easily utilize on-device AI.
Through this Tech Day presentation, we hope you'll gain practical insights and the know-how to take the first step toward implementing AI yourself.
Meanwhile, applications for ' 2025 e4ds Tech Day ' can be made on the official website ( https://www.e4ds.com/conference/techday/ ).
