Former UTD computer science student Viswa Kotra and University of Michigan graduate Pranav Varshney co-founded Nodus, an AI company specializing in compute.
Nodus runs AI computes, an often expensive and tedious infrastructure that guides AI machine learning models, for different clients from models to frontier labs. According to Kotra and Varshney, Nodus helps find the best place to train the AI, find pricing, compile capacity data and figure out potential issues with a model’s training.
The market for such a product is larger than it appears. Kotra likened Nodus to a router for a GPU. For example, those training AI models may not want to deal with finding the best cloud storage system or sift through providers while still wanting to have the capacity to work with a rich training process.
“The question becomes, how can we get the best price for the compute that we need to run all these different AI models?” Kotra said. “We reroute your compute needs to whoever can get it done for the cheapest price.”
Nodus received investment from YC, a startup accelerator. YC works with startups for three months and provides support and connections with other partners like OpenAI and AWS.
Kotra said that his experience at UTD gave him plenty of resources and connections to create Nodus. In his freshman year, he was involved in many clubs like Fintech UTD, the Artificial Intelligence Society, UTD Quant and UTD’s Association for Computing Machinery Research team, known colloquially as ACM Research.
“ACM Research was actually particularly helpful in helping me connect with upperclassmen who, at the time, were aiming for internships or focused on jobs,” Kotra said. “It taught me a lot about the industry and, obviously, the other angle to research the world of AI and machine learning.”
Kotra met his co-founder Varshney during an internship at Scale AI in San Francisco. The internship was Kotra’s first experience in the compute industry. The two continued talking after the internship and kept up with issues revolving around the GPU market.
Both co-founders had experiences at different internships involving AI servers, where they realized that different providers have different GPU offerings, different prices, different reliabilities, and that there was no cohesive way to organize and understand the best route for compute workloads.
“We continued to talk about the different startup ideas when I was interning at Snowflake, and my co-founder was interning at Uber,” Kotra said. “We took the ideas we had from [experiences at] Snowflake and Uber and we basically came to the understanding that there should be a sort of router across this fragmented market, and that was really the birth of this idea.”
Nodus is Kotra’s third startup as he previously created a social media app in high school and later a cryptocurrency credit card company during his freshman year at UTD. He said building a startup comes with daily rejections, mishaps and errors.
“If you don’t have really, really strong conviction, then it’s gonna be hard to survive,” Kotra said. “And I think that has to come from within.”
Down the line, Kotra hopes to expand Nodus’ clients to researchers at universities like Berkeley, Stanford and even UTD.
“I look back on all of the clubs, the friends that I made, the professors that were mentors to me and I believe all of this has led up to me working on Nodus,” Kotra said.
Currently, he plans on taking a gap year, something Kotra says he believe will allow him to focus on Nodus, though he may end up returning to UTD to finish his undergraduate degree at some point in the future.
“San Francisco is great, and if you’re building startups, this is the place to be,” Kotra said. “I encourage other UTD students to also come to San Francisco, try that for themselves and see how it goes.”