AI Chatbot Offers Better Way to Search Maze of Company Info

St. Louis startup is developing a chatbot that can access online company resources.
by Scott Martin

Your company’s internal online directory of resources resembles a medieval labyrinth. Finding something like the in-house holiday schedule can entail hitting lots of walls.

Startup aims to help navigate around those headaches.

The St. Louis company is developing a chatbot to find company information, intending to cut back the pain of lengthy, often fruitless searches. You can talk to Jane like any colleague on your Slack, Skype, email, SMS text or webpage.

Jane’s name was chosen to feel as familiar as those of people in cubicles near you, so that communication with the chatbot would feel natural, according to the startup.

There’s a good shot this could help enterprises everywhere. That’s because Jane’s AI — developed on NVIDIA GPUs —  can comb through apps, documents and other bits of information in databases to help people unearth company answers in moments. was co-founded in early 2017 by David Karandish and Chris Sims, the former co-founders of The 45-person startup recently scored $8.4 million in a Series A funding round.

The company offers a cloud-based service, with subscriptions based on the number of users. So far, it has about a dozen customers, among them utilities, mortgage and financial companies, consumer packaged goods, and universities.

Jane’s Next Act?

Plans for Jane are to go beyond just answering queries. The chatbot can already be used to perform actions across a wide spectrum of enterprise apps and APIs, including scheduling a meeting, creating a ticket, searching your files, returning CRM account details and fetching data from a spreadsheet.

In the future, it might be used to offer proactive messages to take actions, such as reminding you to participate in open enrollment for benefits if it spots that you haven’t already.

Jane’s natural language processing was built on several different deep neural networks and machine learning algorithms to provide the results the company wanted, Karandish said.

Jane Likes Algorithms

“Some algorithms do fantastic under certain scenarios and not so on others — by running them together you get fantastic results in a relatively short period,” Karandish said.

The team at Jane trained its neural networks on NVIDIA GPUs on AWS. The natural language processing does topic clustering and text clustering to help develop pools of answers to questions, and then combines that with entity tagging, part-of-speech tagging and sentence similarity to zoom in on each user’s unique intent.

Jane has “a whole ensemble of different models that are running in real time in production on GPUs,” said Dave Costenaro, the company’s AI lead.