Sit, stay, deploy: Lessons from a real-world robotic blueprint on scaling edge computer vision

This past March, the community gathered for Red Hat OpenShift Commons in Amsterdam. As the "zero day" kickoff to KubeCon and CloudNativeCon Europe 2026, there was no shortage of innovative presentations and discussions, but one session stood out, "Sit, Stay, Deploy: Teaching a Robot Dog with Red Hat OpenShift."

Figure 1., ITQ staffers Johan van Amersfoort, Chief Evangelist and AI Lead, and Sander Harrewijnen, Technologist speaking at the OpenShift Commons Gathering in Amsterdam.

Presented by ITQ staffers Johan van Amersfoort, Chief Evangelist and AI Lead, and Sander Harrewijnen, Technologist, the session moved beyond theoretical AI to showcase a practical—and adorable—application of cloud-native technology.

The challenge: Speed versus complexity in the AI era

van Amersfoort opened the session by highlighting a critical bottleneck in modern enterprise IT, the time to value. AI is developing at unbelievable speed, with breakthroughs in multimodal models and agentic AI happening monthly, and organizations just can't afford long lead times.

"We know for a fact... that building an opinionated Kubernetes platform can easily take 18 months," van Amersfoort noted. "The reality is simply that a CIO can’t wait for 18 months."

ITQ, an IT solutions company, and provider of IT services like Private AI, set out to prove that by using the right platform, organizations could move from 0 to 60 in a matter of weeks rather than years. To demonstrate this, they introduced Q9, a robotic dog that they planned to transform from a "personality-free" machine into an AI-powered office companion.

Figure 2., ITQ private AI platforms

The solution: A private AI platform powered by Red Hat OpenShift AI

The ITQ team needed a robust environment to handle the intense computational demands of training AI models while maintaining the flexibility to deploy at the edge. Their solution was a private AI platform built on:

• OpenShift AI: Used as the primary AI workbench and training stack and to build the conversational AI service for the robot dog.
   
• NVIDIA and Dell infrastructure: Using Dell systems powered by NVIDIA GPUs in their own data center to handle model training and NVIDIA NIMS for model serving.
   
• Kubeflow: Fully embedded within OpenShift AI to manage the machine learning (ML) lifecycles.

By standardizing on this "AI factory" reference architecture, ITQ was able to rapidly develop 2 distinct AI use cases for Q9. This full stack enabled the ITQ team to build up the environment to train and deploy the AI-powered application in a matter of days instead of months. 

1. Computer vision: Training hand gestures

To make Q9 "listen," the team taught it to recognize hand gestures. Using a dataset of roughly 30,000 pre-labeled hand images, they used the YOLO (You Only Look Once) model as a base.

Through multiple training iterations (epochs) on Red Hat OpenShift, they improved the model's accuracy from a barely usable 75% to over 90%. The result? Q9 can now recognize a "hello" or "sit" gesture in real-time, even when the model is running locally on a laptop at the edge.

2. Conversational AI: Giving Q9 a personality

Beyond simple commands, the team used an open source large language model (LLM), Llama 4 Scout, served via OpenShift in their data center. By providing a "system prompt," they gave Q9 a unique personality. Q9 views his ITQ human counterparts as "biological admins," and it is capable of analyzing the room through its camera, generating a witty LinkedIn post about the event, and interacting with the world with a touch of AI-driven sass.

Evidence of success: Scalable and efficient AI

The most compelling evidence of the solution's effectiveness was the live demonstration. While the training required the massive power of the OpenShift AI platform and GPUs, the final optimized models were efficient enough to run on a standard laptop.

This shows that OpenShift AI is able to provide the "essential plumbing" needed to bridge the gap between heavy, state-aware AI training and lightweight, real-time edge inference.

Ready to accelerate your AI journey?

ITQ’s "Project Q9" is a blueprint for how enterprises can use OpenShift to modernize infrastructure and adopt AI with speed and confidence. Whether you are looking to escape costly legacy virtualization or want to build a foundation for the next generation of AI applications, the path forward is clear.

Are you ready to see what your organization can achieve when you sit, stay, and deploy with the right platform?

Want to see Q9 recognize hand gestures and interact with the crowd in real-time? Watch the full demonstration from Red Hat OpenShift Commons Amsterdam to see how ITQ integrated YOLO models and LLMs with the OpenShift AI platform.

See how to use OpenShift at the edge. Run your applications on edge devices, on-premises, or in the cloud.

For an example of another vision model use case, with checks for safety equipment like hardhats, vests, masks, etc, check out the multimodal compliance AI quickstart.