OneCup AI Builds Livestock Face Recognition and Monitoring Platform with NVIDIA

OneCup AI provides Bovine Expert Tracking and Surveillance (BETSY), an artificial intelligence (AI)-powered, facial recognition, tracking, and surveillance platform for the ranching industry. By layering Amazon Web Services (AWS) on top of NVIDIA’s solutions procured from AWS Marketplace, OneCup AI was able to build, scale, and run hundreds of experiments. This enabled the company to create AIs that fluidly process video footage of livestock.

In addition, BETSY had to be able to process the estimated billion data points produced each month by individual ranches situated in rural environments. “Internet performance in rural parts of the world is very poor,” Shmigelsky notes. “In our core market of Canada, the average upload speed is only two megabits per second (Mbps) while the average download speed is 10 Mbps. To move data into the cloud for processing, two Mbps is not going to cut it. A single BETSY camera runs at four Mbps, meaning four cameras will saturate a ranch’s maximum speed of 15 to 20 Mbps.”

According to Shmigelsky, “We took DeepStream and extensively modified it to do what BETSY’s 25 separate but linked AIs needed, extending the standard two layers to five. As each AI layer does something different, we had to train, build, and customize each with specialized datasets, architectures, and medical parameters. We also don't just build one AI that goes everywhere; we build a custom identification model for each ranch. There was no way to scale our technology up and test it without accessing DeepStream’s prebuilt systems in AWS Marketplace and combining it with models for training. It simply was not viable. With these solutions, we can now scale worldwide quite easily."

To tackle the data itself, OneCup AI looked to two additional AWS solutions. As a language for application programming interfaces (APIs), GraphQL enabled the company to query and manipulate data easily through an intuitive and flexible syntax while the fast, scalable, and serverless Amazon Timestream series database service for internet of things (IoT) processed all the data. Shmigelsky notes, “That's how we’re able to handle billions of records.”

AWS services and NVIDIA solutions procured fromo AWS Marketplace enabled OneCup AI to gain the scalability and speed needed to surmount the data challenges of livestock recognition. Shmigelsky explains, “Layering AWS services on top of NVIDIA’s solutions enabled our company to build, scale, and run hundreds of experiments. We ran these experiments on different model architectures, as well as training sizes and meta parameters. All to create AIs that are simply fluid when it comes to processing video footage of animals.”

“Starlink will enable us to run BETSY at full power at millions of ranches around the world, using the AWS Cloud,” Shmigelsky points out. “Consequently, we will be able to process and consume information on a scale that was previously unimaginable in the agricultural technology industry.” With internet data limitations lifted, OneCup AI expects to expand its use of AWS services to continue to develop and extend its solution offerings.