Startup Glean has built a software work assistant with intuition to help its users find exactly what information they need when they need it.
They could want to know how, for example, to access data in a SaaS app used by their employer but there is no user manual and the app’s structure isn’t that friendly. They may have to access Slack conversations and text documents scattered around their enterprise’s distributed systems. Glean has built a system, which uses generative AI, that can literally glean information from multifarious distributed and different sources, collate and present it to users.
Glean CEO and co-founder Arvind Jain told us in a briefing: “We’re established with lots of large clients and customers” including Confluent, Databricks, Grammarly and Okta.
Glean was started in 2019 by Jain and three others who are all involved in engineering: T R Vishwanath, Piyush Prahladka and Tony Gentilcore. Jain was a co-founder at data protection and security powerhouse Rubrik, where he ran R&D. The starting point that triggered the formation of Glean was Jain’s realization at Rubrik that his team members shared a common irritation; finding information about things needed to do their work was difficult.
As Rubrik’s breakneck growth increased the R&D headcount to more than 1,000 people in four years, Jain said productivity did not rise commensurately. Internal surveys revealed that people could not find information needed to do their work. SaaS apps were a significant contributor with, according to Jain, more than 3,000 in use across Rubrik. Information content was highly fragmented across these systems.
Jain realized that the problem affected many businesses and other organizations; it was horizontal and there was a framework that could be developed to solve it. Broadly speaking, that involved building connectors for each API-accessed data source that understood the source data formats and could copy data. Ingested data would then be stored in a standard format in a single data source so that people accessing Glean could find it.
Glean started developing its software with $15 million A-round funding in 2019. Early product development was solid enough to get $40 million B-round funding in 2020 to build out the business. Just a year later came a substantial $100 million C-round to grow the business energetically. The VCs are certainly convinced Glean has a solid offering with lots of potential.
The software builds relationships between information items when it detects connections between them. Jain said: ”We have a standard connector framework.” About a fifth of Glean’s R&D effort is focused on connectors with the bulk devoted to working out how to understand the relationship between gleaned data items. Gleans wants to link, for example, information about a customer in a text document, a spreadsheet, a Slack conversation and sales-related SaaS apps. It needs to understand that these four information sources refer to a single customer and so be able to answer user questions about that customers.
Suppliers such as Hammerspace orchestrate information. Glean said Hammerspace has connectors into file systems but not SaaS apps.
Some of the knowledge will be domain specific, relating to, for example, aviation, legal or medical domains. Understanding the domain and its terminology will clearly be a great help. Glean says it can bring all of a company’s knowledge together for easy discovery and navigation.
Users access a central work hub, with an announcements carousel at the top of the home page to keep up to date with their organization’s news and search for information. There’s a directory function so they can search for and understand who people are, what they’re working on, and how they can help. The people can be filtered and found by department, role, and location. Glean is a way of surfacing, for example, all relevant HR information and enabling a manager to produce regular reports on what their team members have accomplished.
Glean is fundamentally building complicated and fulsome metadata about data items stored in all the different and distributed stored data sources it can access. It’s using large language model (LLM) technology to understand the relationships between the items, take in queries about these items, and produce responses about them. It’s founded on expertise in search and has extended this into new directions using LLM technology, becoming a work concierge chatbot and data store.
Read a Glean FAQ to find out more.
Bootnote
Soham Mazumdar, another Rubrik co-founder, is also active in startup land, being involved with an AI-enabled data analytics startup which is still in stealth.
