ViSenze Powers Visual Shopping Experiences with AWS

To further support its expansion, ViSenze has been able to strategically control costs by running more than 50 percent of its infrastructure on Amazon Elastic Compute Cloud (Amazon EC2) Spot Instances. Spot Instances are used for both asynchronous APIs/noncritical workloads as well as real-time APIs/mission-critical workloads. For noncritical workloads, ViSenze’s search engine can process millions of images in an hour. “There has never been a time when we could not get enough Spot Instances to process our noncritical workloads,” says Xu Fan, principal engineer at ViSenze.

For enterprise customers’ retail apps, the platform runs heavy workloads with deep- learning models and must return search results within 1 second. Engineers have been able to configure their APIs to effectively extend their AWS Spot Instance termination notice from 2 to 10 minutes, which gives the platform enough time to shift workloads to On-Demand Instances as needed and shift back to a Spot Instance as soon as one becomes available.

Using Kubernetes is key in the orchestration of workloads on Spot Instances, because it consistently optimizes the environment to keep costs minimal. Most of ViSenze’s architecture is containerized, which empowers the team to consume more Spot Instances through automation of data ingestion. The team uses Amazon Elastic Kubernetes Service (Amazon EKS) to manage and scale more than 20 specialized microservices.

In addition to its core AWS architecture, ViSenze has recently begun using managed services such as Amazon X-Ray to monitor latencies between microservices. In accordance with its service level agreement with customers, 95 percent of results for real-time APIs are returned in less than 1 second, even for new product images. Search results for existing product images are returned in under 200 milliseconds. “AWS is a truly comprehensive environment offering computation, database choices, hardware compatibility, and new technologies,” says Xu Fan.

Speed is important not just for the company’s customers, but also for internal operations. The startup uses AWS Enterprise Support daily to resolve issues in the most efficient manner and avoid any downtime. “This is a must-have because we have cases that require very fast support. It’s necessary for our engineers to have priority support so our tickets are resolved quickly,” Xu Fan says.

In 2018 alone, ViSenze grew 200 percent year-on-year, and the seamless scalability of its platform is a hallmark of the product. With AWS, it has expanded to support billions of images and users daily. The startup was one of the first companies to start selling ML solutions in AWS Marketplace in December 2018 as an AWS Partner Network (APN) Advanced Technology Partner. Other companies can now leverage ViSenze’s models on Amazon SageMaker within their own AWS accounts.

Adam Monaghan, global marketing manager at ViSenze, says the team is also looking to become an AWS Machine Learning Competency Partner. “This will take us to the next level to be recognized even more credibly across the AWS ecosystem and its clients,” he explains. “It’s obvious that everyone at AWS is keen to see us grow, not just for the sake of AWS but primarily for us as a startup client. They do this by truly understanding what our long-term goals are and how they can directly contribute to our success.”