Sai Guduguntla was 21 when he started doing research at UC Berkeley's RISE Lab - an autonomous robotics project called ERDOS that sent data streams through robotic systems in real time. He was already building for 12 years by then. A self-taught full-stack engineer who picked up chess at the same age he picked up code. He was also singing in the Men's Octet, one of Berkeley's oldest a cappella groups, recording covers in English, Hindi, Telugu, and Tamil on YouTube. None of this looked like a startup story. All of it turned out to matter.
The summer before his senior year, he was at Salesforce, working on Einstein AI Chatbots. The summer after, he was a founding engineer at Bloom (YC W21), a financial education app built for people just starting to invest. He saw what it meant to build at the pace and stakes of YC. He graduated in 2022 and immediately co-founded Calypso, a fintech platform for loan servicing. Then he met Atul Raghunathan - an ML engineer from Meta who had spent years building the models that decided which ads 3 billion people see.
Together they applied to Y Combinator Summer 2023 with an idea for AI that would automate email outreach. The premise: replace SDRs with software. The reality check arrived fast. You cannot fully replace a human in a complex B2B sales conversation with the technology that existed in 2023. So they pivoted. Instead of replacing reps, they asked: what if you could make every rep dramatically better?
The answer started with a very simple, very embarrassing observation. Sai and Atul had no sales skills. They were pitching enterprise buyers and losing. Rather than hire someone to pitch for them, they built an AI buyer - a simulated prospect with realistic objections, a persona, and a voice. They practiced with it. Then they let other people try it. The response was instant. Reps called it "the next best thing." Managers called it the coaching they never had time to give. IBM and LinkedIn called their account executives.
Before Hyperbound launched publicly in January 2024, Sai had conducted over 2,000 user interviews. Not demos. Not surveys. Conversations. "We didn't start with a great product," he said in a Predictable Revenue podcast. "We started with a lot of conversations." The result of those conversations was a product that analyzed real sales calls, identified the behaviors that close deals, and turned them into customizable AI roleplay scenarios a rep could practice any time, at any scale. Autodesk, Bloomberg, Vanta, G2, Monday.com followed.