INTRODUCTION
Intelligent identity and security management for businesses
In an ideal world, our devices would instantly recognize us and grant access to our apps and documents — while safeguarding our information from everyone else. When networks were private and staff worked in the office, security was relatively easy for organizations to manage. But now, with the shift to the cloud and the rise of remote work, security and identity management are much more complex.
As the world’s leading identity security provider, Okta aspires to free everyone to safely use any technology on any device, all powered by their identity. With more than 18,000 customers globally, Otka has developed integrations for more than 7,000 applications.
To advance its identity governance and administration efforts, Okta acquired the workplace operations platform, atSpoke, in 2021. Rebranded as Okta Inbox, it makes requesting access to applications easy. For example, Okta Inbox’s built-in artificial intelligence means users can input natural language queries to find what they need quickly and easily.
“We’re customer-obsessed. It’s in our DNA, from senior management to the development team. Everything we do is designed to help customers work faster and stay secure,” says Suchit Agarwal, director of engineering at Okta.
THE CHALLENGE
Finding a scalable way to power natural language queries
To create Okta Inbox, Okta wrote code and manually built machine learning models to power natural language queries. These models generate vector embeddings, a series of weighted values that detect a data set's qualitative and quantitative aspects.
While the approach was effective, it needed more scalability as the business grew and limited how much the team could experiment. For example, when searching a vectorized collection of apps using natural language, the machine learning model would recognize that terms like ‘video conferencing’, ‘make a video call’ or, ‘conference call’ relate to that company’s preferred communication app.
Okta’s machine-learning models and embeddings were managed and maintained on a dedicated Kubernetes cluster. However, vectors either need to be stored in a dedicated vector database — which is an additional cost and solution to maintain — or a bolt-on search engine with vector search functionality needs to be installed. Adapting the environment to increase performance would incur additional fees and take time away from developers.
THE SOLUTION
Automating embeddings and optimizing algorithms
When given a first look at MongoDB Vector Search on Atlas, Okta realized that automating the manual work underpinning natural language queries with the solution would allow developers to focus on innovation.
“MongoDB Vector Search on Atlas was the answer to our problems. It simplifies a lot of the work that goes into making Okta Inbox super user-friendly for customers,” said Agarwal. “It was amazing to find a solution that integrates so easily with MongoDB Atlas. We didn’t have to compromise on having a single data platform.”
Instead of building a custom solution for embedding storage, embeddings could now be stored in MongoDB Atlas in collections. When prompted, MongoDB Vector Search on Atlas queries embeddings to find the right app quickly and easily. With the previous approach, Okta relied on home-grown search algorithms for these types of queries. MongoDB Atlas provided one efficient data layer that simplified these operations.
“Okta Inbox knows which app the user is requesting access to and automatically raises a ticket to process the request,” Agarwal explained. ”It receives thousands of requests through our chatbots, and we no longer need to manually process thousands of text embeddings to resolve those queries.”
.svg)