AI & Machine Learning solution consulting and administrative maintenance of QuarticON infrastructure

QuarticOn is a technology company that provides online stores with modern solutions in the Software as a Service (SaaS) model to improve their operations by automating marketing activities and personalizing the available offer. The company's main asset is its state-of-the-art technology, which is based on the use of artificial intelligence and machine learning algorithms, enabling services to operate in real time and with minimal human resources.

Needs and cooperation

In 2021, QuarticOn received funding from the National Center for Research and Development (NCBiR) in the amount of PLN 4.7 million (total project value of 6.3 million) for the implementation of the AI Fashion Stylist project. Part of this amount was used to migrate the company's resources to the AWS cloud and to implement the necessary tools that fit into the MLOps approach.

As part of the project, we are working with the client providing consulting and implementation services, including the configuration of an EC2 P4d instance equipped with NVIDIA A100 Tensor Core GPUs, which is designed for machine learning-based workloads. These instances are dedicated primarily to analysts and developers who develop Machine Learning-based technologies. QuarticOn is using them in its Fashion Stylist AI project, in deep learning AI on some 5 million clothing images. Based on these, the neural network learns how to compose sets of clothes from the available assortment to create a cohesive look for different occasions. The AI then identifies the collected styles and acquires knowledge about current trends. The challenge for the instances used is the very high load resulting from the rare huge number of parameters of the analyzed images, i.e. the varied, often very high resolution of the images, which are represented as 3 pixel arrays for each RGB channel. The entire learning process involves optimizing the network structure by performing continuous experiments. One learning cycle, or "walk-through" of all the images, is several days. Due to cost optimization, Instances are run on demand in different configurations, only when another network optimization process is scheduled.

The second part of the client's infrastructure is auto-scalable resources, responsible for the high availability of the application, which delivers the results of Fashion Stylist AI analysis to Quarticon's clients' online stores.

Summary

Our cooperation has resulted in the optimization of infrastructure maintenance costs, and the consulting services we provide have contributed to the smooth implementation of solutions dedicated to machine learning. In addition, we provide QuarticOn with 24/7 administrative support services for the entire infrastructure.

ANY QUESTIONS? CONTACT US

Read also:

• Building Kubernetes-based infrastructure and taking care of resources
• Implementation of cloud infrastructure for Magic Commerce, based on microservices, containers and IaC approach
• Solving problems and increasing efficiency for Oferteo
• Migrating and building infrastructure in AWS for SimpleMining.net
• Building infrastructure in IaC approach for Future Point