Nutanix’s system engineers (SEs) are responsible for pre-sales services. When customers turn to the company for infrastructure, SEs are the ones who build out solutions based on their specific needs. So, when these engineers have questions about what is possible, these questions are usually complex, nuanced, and highly specific to a given customer—making it a special challenge for Nutanix to create an AI solution aimed at providing instant answers.
“A common phrase in the SE community is that, for most questions, the answer is: It depends.” said Ratan Kumar, director of product management for SaaS engineering at Nutanix.
Kumar is the project manager for SEGPT, a purpose-built AI solution for Nutanix SEs.
“The SEGPT app cannot say: It depends,” he explained. “It has to provide an answer. And it has to provide the right answer.”
With more than 700 SEs across the company, questions can range from basic product specifications to intricate technical integrations. For instance, an SE might need to know which network cards are supported by a specific hardware platform, understand resource allocation requirements for disaster recovery scenarios, or simply explain to customers why Nutanix solutions are a better fit for a specific use case than the infrastructure offered by competitors.
While other AI tools might struggle with such specialized queries – or, worse, provide plausible sounding “hallucinations” – Nutanix officials wanted to build a system that could provide answers with the same level of authority as the company’s most experienced engineers.
“If you ask SEGPT how many planets there are in the solar system, it will say it doesn’t know, because it doesn’t have the data,” Kumar explained. “But you can ask it anything about Nutanix, and it will answer beautifully.”
That is the domain context we have brought to the purpose built Gen AI solution.
Building and Training a Custom Model
Building and using GenAI apps for business productivity has become a big trend.
The 2025 Enterprise Cloud Index report by Nutanix, which 1,500 IT and business decision-makers worldwide, found that nearly 85% of responding companies already had a GenAI deployment strategy in place and 55% were actively implementing it. Only 2% said they hadn’t yet created a plan for GenAI deployment, and none said they had no intentions of creating a GenAI strategy.
Having the right data, organized and accessible is an important step in the GenAI app development process. Jirah Cox, field chief technology officer for the Americas at Nutanix, helped track down the information needed to answer complex customer queries.
“It can feel like looking for a needle in a haystack,” he said.
“Sometimes our administration guide for a product could be hundreds of pages long. Figuring out how to do one procedure is a big searching effort. If I can just ask SEGPT how to do something, and it tells me the three steps I need to take, that’s a dramatic time saver.”
The tool also reduced the extent to which SEs need to rely on the sort of institutional knowledge that is often diffused among various senior employees at the company – or, worse, scattered across countless different Slack threads.
“With my tenure and experience, SEs naturally want to consult with me on deals and configurations,” Cox said. “That’s great, but it’s also really wonderful for the SEs, and for me, to have another place to turn to get the right answer.”
To build out and train the SEGPT model, Nutanix engineers coordinated with SEs to track down the “sources of truth” that they rely on in their jobs, Kumar said. After indexing that information, the SaaS team trained the model across different data formats, including PPT, PDF, Word, and Excel files.
“We ran a pilot in which we asked the model questions, and then we would run those answers by the SEs to make sure the bot was responding correctly,” said Kumar.
“Then we’d fine-tune the model. It’s a continuous process of evolution. The GPT becomes more and more intelligent.”
Cox praised the ability of SEGPT to not only return accurate, actionable answers, but also to provide documentation to validate those answers.
“What’s great about the tool is that, along with a summarized response, it links back to the admin guides and tech notes where it found the information,” he said. “That’s very different from our experience with a lot of the commercial AI platforms.”
The Road Ahead
The Nutanix SEGPT, SalesGPT and SupportGPT apps benefited from the pioneering work that went into building SupportGPT, according to Kathy Chou, senior vice president of SaaS Engineering at Nutanix.
“They use the same code base, which is what makes our xGPT efforts at Nutanix scalable,” Chou told The Forecast.
It took us about 10 months to create SupportGPT, the first GPT app her team built. Those early efforts required a lot of testing, benchmarking and refinement.
“Since then we have been able to ramp SEGPT and Sales GPT in half the time,” Chou said. “We are easily able to leverage the technology to different functions as we can train the system for different data sets. Our goal is to build a standard agentic framework and transition to NAI - Nutanix' Artificial Intelligence product so our external customers can also benefit from what we learned internally.”
The ultimate goal of the SEGPT initiative, Kumar explained, is not merely to make SEs more productive, but to give them time back that they can spend working closely with customers.
“We expect to see customers who are more satisfied, because our folks are able to spend more time with them and understand more of their problems, rather than having to spend so much of their time in-house, tracking down information,” he said.
Nutanix is still in the process of rolling out the solution across all SEs, but early results have been promising. In his first several months using the tool, Cox said, he spent only a few minutes per day in the model, spread out across a handful of inquiries. But some of those several-minute engagements are potentially saving him hours in wasted time.
“There are certain questions that lead you down a rabbit trail,” Cox said. “For those, you might have to spend 30 minutes opening 30 different browser tabs to find the pearl you’re looking for. So, for every second I’m in the tool, I’m seeing a 10x or 20x time savings.”
Eventually, the solution may be folded into a larger company-wide GPT solution, along with existing tools for sales and customer support. But for now, Kumar and his team plan to improve SEGPT over time by allowing users to quickly rate the responses they receive—ultimately leading to answers that are even more authoritative and actionable.
“It’s like having a really smart friend who never sleeps, and who is constantly looking at new information,” said Cox.
“I only learn something new when I take the time to check the latest manuals. But the app is scraping the documentation all day, every day. When I ask a question, I know I’m going to get the latest answer.”
Calvin Hennick is a contributing writer. His work appears in BizTech, Engineering Inc., The Boston Globe Magazine and elsewhere. He is also the author of Once More to the Rodeo: A Memoir. Follow him @CalvinHennick.
Ken Kaplan contributed to this article. He is Editor in Chief for The Forecast by Nutanix. Find him on X @kenekaplan and LinkedIn.
© 2025 Nutanix, Inc. All rights reserved. For additional information and important legal disclaimers, please go here.