Standard Cognition develops AI-powered checkout experiences for brick and mortar retailers with its computer vision platform. In this talk, Standard Cognition’s Tushar Dadlani discusses the challenges in delivering a no-touch checkout experience that responds to constantly changing data in real-time.

At Standard Cognition, streaming vision-data is converted into high velocity and variety metadata that requires purpose-built databases for developers to prototype on the data and build AI models quickly. The team faced several challenges in building applications on vision data and ultimately moved away from traditional databases and towards a real-time database in the cloud for the following reasons:

• Velocity of vision data: Standard employed three data strategies to handle the velocity of vision data: keeping high frequency (~ 500 Hz) data on the store premises, immediately processing low-frequency (~ 5Hz) data in the cloud and streaming medium frequency (~ 50 Hz) data for application development.

• Variety of vision data: The schema of the metadata changes on a daily basis which resulted in the team selecting databases that could handle frequent schema changes, mixed data types, and complex objects.

• Developer friendly: Developers needed to rapidly iterate and build data powered applications on production datasets and so Standard selected databases that made the data easily available via REST APIs.

Tushar shares a data stack that handles data at different frequencies, from different sources and is easily accessible to developers for ad-hoc analysis, prototyping, and moving new features into production. Standard uses Rockset, a real-time database in the cloud, for speed and simplicity when developing new features and verifying the accuracy of their AI models.