With clients such as the US Department of Defense and global automotive manufacturers, VizSeek likes to think of itself as a small company with big customers.

Launched in 2015 and a member of the Oracle for Startups program, VizSeek is a visual search platform that uses artificial intelligence to find images and drawings of a company’s manufactured parts and components based on their shapes. Most VizSeek customers use its platform to find work they’ve already done, matching photos to existing engineering drawings, or instantly identifying needed replacement parts. For makers of ships, cars, and planes, it’s an extremely useful capability that can translate to less wasted time and increased profit margins.

The concept of visual search has been around for years, popularized by consumer services like Google Images and Pinterest Lens. In contrast to web search engines, VizSeek creates a searchable database of a company’s proprietary digital assets that’s both restricted and secure, yet highly available and searchable by authorized employees.

Retailers, manufacturers, and engineers can then take advantage of this searchable database to find files by shape. This helps manufacturing staff at all levels, including aftermarket parts sales, and enables them to upload photos, 2D or 3D CAD models, or text to instantly locate products, along with the data that goes with those products.

Put simply, VizSeek helps employees use visual search to find products and designs in files they don't know the name of, but in some way contain a shape. VizSeek, which runs its AI-based platform on Oracle Cloud Infrastructure (OCI), is unique in its ability to cross platforms and file types.

“Our customers can drag and drop an Excel file into VizSeek, and it will find related data that's connected with what they’re looking for,” says Rob Hill, chief architect for VizSeek. “They can drop a photograph, and it will pull up a 2D drawing. They can drag and drop a PDF, and it'll pull up a 3D model. It can cross all of these different platforms.”

In a market expected to reach nearly $15 billion within the next 12 months, VizSeek needed a faster way to scale.  After migrating its AI-based platform to OCI and running its AI visual search models on Nvidia V100 GPU cloud servers, “We've actually seen a performance increase close to 50% compared to on-premises, which is awesome for our customers,” says Hill.

VizSeek’s target market tends to be companies in the engineering, ecommerce, and customer service industries, with a particular focus on product design, manufacturing, quoting, and the aftermarket sale of parts. For example, by using VizSeek’s visual and shape search, a design engineer can locate parts they’ve made previously without having to recall the exact product name or description required in a text-based search. Similarly, a quoting engineer can upload a photo, 2D hand sketch, 2D drawing, or a 3D model to find comparable existing parts that were previously quoted.

“We have customers where, before, it would take them days to issue a quote,” Hill says. Now, they can input a customer’s file, and VizSeek returns similar quotes within seconds. “The quicker you can turn around a quote, the more likely you'll be to get that deal.”

In the early years, VizSeek gained traction through project work with manufacturers and organizations such as the National Science Foundation and US Department of Defense. More recently, demand for VizSeek’s technology has increased, as businesses both comprehend the value of visual search and warm to the idea of cloud technology to gain a market advantage over competitors.

VizSeek chief architect Hill discussed the architecture behind VizSeek’s tech platform with Oracle cloud architect Cody Brinkman for a recent episode of Built & Deployed, a video series featuring software architects discussing cloud technology.

Powering visual search

Since VizSeek migrated to OCI, customers can now access the company's virtual cloud network using an internet connection to Oracle Web Application Firewall. Users are then passed through VizSeek's firewall, which is secured by Oracle Security Lists and Cloud Guard, to its load-balanced Windows web servers. These servers are set up on two OCI virtual machines, each located in its own availability domain for high availability, with developers entering the network through a VPN.

Once authenticated, these front-end web servers send user search requests to VizSeek's database systems. In addition to using Oracle MySQL Database Service with the high-availability option, VizSeek uses three clustered Redis databases, which are each paired with GlusterFS systems, with each pair sitting in its own availability domain. The search criteria is then indexed on two OCI back-end load-balanced Linux servers that run VizSeek’s proprietary indexing software.

The front-end also communicates with two Linux searching servers through the back-end load balancer. These searching servers communicate with the database systems as well as with VizSeek's GPU servers on OCI. VizSeek runs its AI visual search models on two Nvidia V100 GPU load-balanced servers, each sitting on a VM in a separate availability domain. The AI models use TensorFlow and PyTorch to help extract product metadata from any PDF, presentation, raster image, or 3D model, and then classify the files into searchable digital assets.

“Our software is a visual search engine, and people expect it to be fast, they expect it to be reliable,” Hill says. “And we also need to be able to scale for our customers and make sure our platform is secure. So, this type of architecture really meets all those requirements. We’re able to debug and scale without any error or interruption.”