Pryon, a startup developing an AI-powered platform to look for insights in — and surface answers from — enterprise knowledge bases, today announced that it raised $100 million in a funding round led by Thomas Tull’s U.S. Innovative Technology Fund.
Pryon’s founder, Igor Jablokov, said that the new cash will be put put toward supporting Pryon’s general growth, expanding its 100-person team, growing its presence in international markets and scaling its strategic partnerships. A source familiar with the matter tells TechCrunch that the funding, which brings Pryon’s total raised to $137 million, values the company at between $500 million and $750 million post-money.
Prior to launching Pryon, Jablokov led the multimodal AI research team at IBM. He left to create Yap, a Siri-like speech recognition startup that Amazon acquired in 2011 to jumpstart development of Alexa. (Fun fact: Pryon’s namesake was the code-name Amazon used for the speech engine underpinning Alexa.)
Pryon isn’t a voice assistant. But it is an assistant — of sorts.
Jablokov describes it as a “knowledge fabric” that can interface with a third-party chatbot or channel, ingesting data like audio, images, text and video and converting it into a format that’s searchable and usable by whatever frontend is connected to it.
An analog, Jablokov says, is Kendra, Amazon’s AI and machine learning-powered service for enterprise search. Similar to Kendra, Pryon leverages connectors to unify and index previously disparate sources of information from databases. But Jablokov claims that Pryon is up to 2x more accurate than Kendra, ingests data up to 10x faster and can index billions of documents versus Kendra’s 100,000-document limit.
“Organizations don’t need to migrate their content into the Pryon platform, as it layers over existing systems of record and doesn’t require end-user retraining to author content in a new way,” Jablokov said. “You simply point to a repository and it generates an AI model from the underlying content. If you have legacy content in there, that’s OK, since Pryon uses computer vision, optical character recognition and handwriting recognition to understand what’s in there.”
Jablokov claims that it takes less than a second for Pryon to create, update or delete content on the platform in a privacy-preserving way — and that the platform leaves no trace of its indexing work.
“Since the customer defines what goes into Pryon in terms of public, published, proprietary and personal data, there’s always attribution to authorship and ownership, so that only content they’re legally entitled to is what’s in there,” Jablokov claims.
Pryon has competition from the aforementioned Kendra as well as Microsoft SharePoint Syntex, which draws on knowledge bases to cobble together answers to company-specific questions. Startups like Hebbia, Kagi, Andi and Glean also tap machine learning models to return specific content in response to queries (as opposed to straightforward lists of results).
But Pryon appears to be doing quite well for itself, notching annual recurring revenue in the “seven figures” and securing “a dozen” large enterprise and public sector clients, including Dell, Nvidia and Westinghouse.
“Pryon is one of the few AI-native companies that was designed for enterprise use from its founding days,” Jablokov said. “It can meet the needs of the most regulated of environments, from energy to government, because of the unique way the platform safeguards content.”