Cloud Infrastructure for Training AI/ML Models

How it started: Seeking an infrastructure partner for AI product development


Fuller Vision is developing an innovative and highly complex product with the mission to literally transform the way we see the world around us. Their product, called Autofocals, are autofocus glasses that provide perfectly sharp vision at different distances. This is achieved through a combination of sophisticated technologies, such as pupil detection software and gaze distance estimation. By analysing eye movement in real time using a range of delicate parameters, the software adjusts the lens prescription based on where the user is looking.

Since the product is based on AI/ML technology, Fuller Vision needed reliable infrastructure to train their AI/ML models.


“We were looking for an agile partner who would allow us to start with a smaller scale project and then progressively develop it through active communication and exploring our needs,” explains Vladimir Vojvodić, Operations Director at Fuller Vision.

After evaluating various providers, the company chose Mainstream’s AI-ready cloud infrastructure, which stood out for its capacity, cost efficiency and availability of local support. 

“Mainstream’s infrastructure met our needs in implementation and quality, as well as cost. Compared to global infrastructures we used before, Mainstream offers better cost control. Another advantage is their geographic proximity, which made it easier to initiate communication and build the partner relationship,” says Vojvodić.

Innovation challenges and how the cloud helps 


Developing products like autofocus glasses is a complex process with a plethora of challenges.


“Our eyes are incredibly delicate organs, and their proper functioning is essential to our quality of life. Moreover, each person’s eyes are unique, with different prescription needs that change over time. Meanwhile, the movement of the eye and our efforts to focus on objects at varying distances produce measurable and comparable data, says Vojvodić, adding:”What further complicates matters is the need for vision interaction and feedback in real-time, within tens of milliseconds. Combine this with the complex mathematical operations involved in analysing eye movement, and you get a problem that can only be solved through neural networks.” 

For training their AI models, Fuller Vision required precision, reliability and speed combined with the ability to run on hardware with extremely small dimensions and a limited energy profile.Mainstream’s cloud infrastructure met all these demands, enabling them to repeatedly train and improve their models. The goal is to ensure the model works as quickly and accurately as possible while at the same time remaining as “lean” as possible in terms of operations and energy consumption. 

“Mainstream provided exactly the type of infrastructure we needed. Our AI teams are now able to run multiple parallel experiments to examine which model parameters yield the best results. This capability is crucial and significantly accelerates our development.” 

Key features of AI-ready cloud Infrastructure


After analysing Fuller Vision’s requirements, Mainstream delivered infrastructure aligned with AI development priorities, featuring: 

  • Virtual machines with NVIDIA GPUs for training power-hungry AI models
  • VMware virtualisation that ensures high availability and consistent operations
  • Scalable resources on the Enterprise Cloud platform
  • High-performance storage for large data volumes and fast access
  • Enterprise networking for secure and reliable connectivity

Additionally, our team provided a partnership approach, including consulting and designing a customized setup, full infrastructure management support, 24/7 monitoring, focus on optimal resource utilisation and ad-hoc support. 

Benefits:

  • High-performance GPUs for AI model training
  • Operational agility for product development
  • Scalable infrastructure and high-performance storage
  • Cost efficiency compared to global providers’ infrastructure
  • Continuous support from experienced cloud engineers, with rapid response in critical situations

Looking Ahead


Fuller Vision has exciting and important plans ahead. Their product is in its final development phase, with plans to unveil the new eyewear model at Vision Expo West 2025, the largest trade show for vision improvement technology. Commercially, they have begun receiving initial orders, with the first product batch expected to ship to the US market by the end of Q1 2026.

Fuller Vision plans to continue collaborating with Mainstream, focusing on further product improvements whilerelying on the cloud infrastructure and team supporting their dynamic development.

Discover how Mainstream can improve your business.

Contact us at sales@mainstream.eu or fill out our contact form.

Other case studies

A leading insurance company ensures business continuity and data security with a personalised Disaster Recovery solution 

With Mainstream’s cloud-based Disaster Recovery solution, a leading insurance company enhanced system resilience, reduced costs, and ensured 24/7 data protection and business continuity.

Optimization of Power BI architecture and professional support for creating custom reports

How the Wurth company improved business reporting with PowerBI architecture optimization and custom reporting.

Designing and building a hosting solution for an advanced e-commerce

How Gigatron, a leading e-commerce operator, achieved IT infrastructure stability, scalablity and security with Mainstream Managed Hosting.