Childhood friends to cofounders
Dhruva and his co-founder Hemang met in fourth grade in Mumbai. They were friends the way kids are at that age, part of each other’s world without thinking much about it. Then Hemang’s family moved to a different state after sixth grade, and six years passed with little contact between them.
They reconnected in twelfth grade while figuring out where they both wanted to go to college. Without planning it, they both applied to and were accepted to Georgia Tech, where they ended up as roommates and graduated together in three years.
After Georgia Tech, Hemang moved to New York to work at Meta, where he eventually landed on the post-training team for LLaMA 3 and 4. Dhruva went to the Bay Area for a master’s at Stanford, then to DeepMind, and then to a seed-stage company called Refuel AI, where he joined as a Founding ML Engineer. Although they were separated, the urge to be around other high-agency people kept them in touch.
“I want to be close to high-agency people,” Dhruva says. “And my co-founder is one of the highest-agency people I know on earth.”
In October 2024, Hemang called Dhruva and said he wanted to leave Meta and start a company. Dhruva said no, as he was heads-down building Refuel and preparing for their upcoming growth. Then the acquisition news came: Refuel was getting acquired by Together AI.
When Dhruva heard the news, he knew he likely wasn’t going to stay at Together AI for long. So he called Hemang back, and they got to building.
Intentional idea iteration
They decided what to build the way many founders do: by making a list. “We approached this in a very pragmatic way; we built out a spreadsheet of almost 15 to 20 ideas,” Dhruva says. “We had metrics for each of these ideas and were evaluating each idea using different scales that made sense to us. Different attributes for different ideas. You’re scoring them.”
They had good material to work with. Hemang had spent years at Meta and had real insight into image generation using flow-based models. Dhruva had just come out of Refuel with a close-up view of how large enterprises tried to integrate AI into their IT landscapes, and how messy and difficult that usually turned out to be. On top of that, they were watching what was happening with code generation. Models were getting dramatically better, and problems that had felt out of reach six months ago suddenly felt solvable.
Out of all of that, two ideas stood out. The first was an autonomous marketing agency, a system that could run campaigns on autopilot using flow-based image generation models. That idea is actually where the company name, FlowGen, came from. The second idea is what they’re building now.
Turning down YC and SPC to find the right idea
They applied to YC and SPC with the marketing agency idea and were offered by both, though that wasn’t really the point. “The rationale to apply was twofold,” Dhruva says. “One, to understand how VCs think. At that point, we’d never interacted with VCs. And we’re like, we have no idea about what this fundraising process looks like. And second, in the process of telling a story of what we want to build, we realized were actually also getting clarity ourselves.”
SPC was first. Dhruva went in to meet with the founding partner, who, within ten minutes of their intro call, told him they really liked him and Hemang, even though they weren’t sure about their idea yet. YC came next, and the interview went well. But that evening, Dhruva got a short email asking if he could hop on a call. Diana, the YC partner, was straightforward: she loved both of them as founders, but she wasn’t sure about the idea. She made them an offer anyway. "Come join the batch, iterate for a few weeks, pivot if needed."
But they turned both offers down. Rather than walk into YC already doubting what they were building and burning through batch time trying to find their footing, they decided to step back entirely and find the right idea first. “We saw YC as a very unique opportunity,” Dhruva says. “If you are going into YC, you want to go in with traction and use it as a launchpad. You don’t want to waste your time there in pivot hell. The best way to use YC is to go in with some traction and use the network to expand from there really fast, so that you’re heading towards a strong demo day.”
But he doesn’t pretend it was an obvious call. “There are drawbacks to this. You may lose time figuring it out outside the cohort. You don’t know if you’ll even get an idea outside the cohort. And maybe you’ll get it faster while you’re advising inside YC.” He sat with both sides of it. “I still, to this date, don’t know if this was the right decision,” he says. “But that was the rationale. And I feel like this is a question to which there is no right answer.”
The bet on ERPs
After stepping back from the marketing agency idea, Dhruva kept coming back to what he had seen at Refuel. How messy large enterprises can be when integrating AI into their IT landscapes. The idea they landed on starts with the ERP layer. Today, those systems require large teams of people to operate them. Dhruva sees a world where that changes, where the ERP has no human users at all, and enterprise operations run autonomously.
“A lot of these operations are supported by very large teams of people,” he says. “They must use SAP; they must use these legacy ERPs to actually run their enterprises and operations. We see a world where ERPs are used by agents. Where there are no human users of ERP, they should just autonomously function.”
That’s what FlowGen is building toward. An autonomous agent layer inside legacy ERPs that lets enterprises operate without anyone touching them. “We want to become enablers of the first generation of zero-touch enterprises,” he says. “A world where very large enterprises are run by very few people.”
Dhruva is straightforward about the challenges of breaking into the space. “Our customers are very large enterprises, and they expect to see older, experienced people on the other side,” he says. “They don’t expect young people to be in this business.” Knowing that, they were deliberate about who they brought in. They raised a $600K pre-seed from Pear VC, joined the Pear accelerator, and within two weeks of demo day had a term sheet signed for their seed round led by Spark Capital. “We’re coming in almost as outsiders,” Dhruva says. “But we’ve got insiders on our side now. We’ve got some phenomenal folks helping us build this company.”
The lowest moment isn't always a signal to stop
There was a moment during the Pear accelerator that hit differently from anything else that year. They invested a lot of time and effort in a proof of concept for a large potential customer, doing it for free, but the deal fell through.
What got them through wasn’t a clean breakthrough. It was a night’s sleep, working back through the problem from first principles, and what Dhruva calls unreasonable belief. “I just can’t believe that we couldn’t do this because we’re bad salespeople,” he says, “because that can change very fast.” One lost deal couldn’t be the whole story. So they went back to work.
Working through it together also taught Dhruva something about his co-founder. There’s a quote, he says, about being bearish when the market is bullish and bullish when the market is bearish. “My co-founder is exactly that,” he says. “When things are going really well, he’s asking what can go wrong. And when things are not looking good, he’s really optimistic.” Between them, someone is always steady. “This is phenomenal to have on the other side,” he says, “because I’m someone who’s super optimistic. When things are going well, I can lose track of things. But he does a great job at being paranoid when things are going well, which is actually really important to have.”
The bigger lesson reaches beyond just the two of them. It shows up in every conversation Dhruva finds himself in now, with customers, investors, and people they’re trying to hire. “A lot of the alpha in building a company is in being able to uncover the unsaid,” he says. “The question behind the question. The intent behind their answers, the intent behind their behavior. If you can figure that out, you can really crack everything.” He and Hemang have been building frameworks for it and working to get better at reading what a specific person is actually worried about when they walk into a sales call. “This is a part which we have been really trying to figure out,” he says. "We’ve made a lot of progress.” It’s simultaneously, he says, one of the hardest things you have to do as a founder.
Legacy software, new playbook
Today, they have a team that understands the ERP ecosystem from the inside, enterprise customers that are responsible for running the world, and a clear picture of what they’re building toward. The ERP market is one of the largest and most entrenched in enterprise software. SAP alone has tens of thousands of customers running their entire operations on its systems.
Dhruva and Hemang are betting that those customers will eventually need a layer on top of that which removes the human element entirely. They want FlowGen to be that layer. They’re in the market, and they’re growing fast.