Ashwini Asokan and Anand Chandrasekaran were vacationing at a relative’s home in Bengaluru when the renegotiation came through. It had ruined their time-off. They sat, not talking to anyone, when Asokan pulled out two slips of paper.
Hours earlier, in December 2023, they had been ready to sign a $50 million Series D fundraising round, which they had expected would finally give their Chennai-based artificial intelligence startup, Mad Street Den, room to breathe. But on the day of signing, the investment firm came back with new terms: no secondaries, or stake sales, for the earliest investors; stake dilution that would punish loyal believers; and board-control provisions that would effectively strip the company of its independence.
“It wasn’t bad for us or the team,” Asokan said. “But it was terrible for the people who backed us from the beginning.”
For most founders, this is a poison pill swallowed in silence. Instead, after speaking with their third co-founder, Costa Colbert, Asokan and Chandrasekaran each took a slip of paper and wrote down the same question.
Asokan, in her looping hand: What is left for me to learn? Chandrasekaran, in block print: What is left for me to learn?
They both knew the answer: walk away from the deal and eventually from the company itself.
Two people, one instinct
Post a degree in electrical engineering from the Indian Institute of Technology, Madras, Chandrasekaran acquired a PhD in neuroscience from Texas’ Baylor College of Medicine. Then he headed to Stanford University, where he worked on Neurogrid, a 16-chip board that simulated a million neurons in real time on 3 watts, roughly what a tablet draws. Asokan went from a visual communications course in Chennai to Carnegie Mellon University for a master’s in interaction design, and then Intel.
On paper they both worked in artificial intelligence. In practice they were opposites. Chandrasekaran was the builder, happiest at a whiteboard. Asokan was the translator: she could walk into a room of sceptics and hold it. He worried about whether a system would hold at scale. She worried about whether anyone would trust it enough to try.
Their first clean break came on New Year’s Day 2013 in California. Asokan woke up, turned to Chandrasekaran, and said: “Let’s put the house on Craigslist.” Within 48 hours the furniture was gone and the car was sold. Then they booked their tickets to India.
They were leaving the Valley just as it was beginning to reward them: Chandrasekaran had started work on projects funded by the Defense Advanced Research Projects Agency, the US Department of Defense’s research arm, at a startup near Stanford; Asokan was climbing from one promotion to the next at Intel. Yet they left because it felt right before they could explain it.
Chandrasekaran had a plain read: they did not yet know precisely what they were going to build, but India was the cheaper place to take time figuring it out. The same instinct would run through everything that followed: starting Mad Street Den in Chennai, and shutting it a decade later.
The OG of AI
Mad Street Den, or MSD, was one of India’s first serious AI companies long before generative AI became a buzzword. When Asokan later wrote to early investors to tell them that she, the chief executive, and Chandrasekaran, the chief technology officer, were winding down the company, many replied with the same line: MSD, and its internal platform Vue.ai, had been the OG of AI in this part of the world, a full decade too early.
In 2015, in the middle of a routine cycle of pilots, Chandrasekaran had hacked a vector search engine in about two weeks. It was a fix for a basic problem: how do you let machines search not by exact keywords but by patterns of similarity buried in images, text and transactions? A database might say “red shirt”, but the signal a customer cared about—crimson or scarlet, broad checks or thin—lived in the image itself.
That two-week hack became the foundation of MSD’s architecture. Years later the world would call the pattern “retrieval-augmented generation”, or RAG. “I built something very similar to RAG in 2015,” Chandrasekaran said. “I didn’t know whether it was important, but it ended up becoming the foundation layer for the stack and started powering customer use cases.”
What they shipped was not an app or a chatbot but infrastructure: a three-layer stack with data pipelines at the bottom, the vector search engine above that, and a contextual layer that applications could call.
Explaining it was harder than building it. “I had to write definitions of what embeddings mean, what latent search means,” Chandrasekaran said. “I kept asking myself, how do I say this in English?” So they stopped explaining and started showing. A retail catalogue sorted itself. A loan application moved through a queue in seconds. A visitor walked into an office building and the system knew who they were. What got enterprises to sign was never the architecture. It was watching AI do something.
India had no playbook for enterprise deeptech startups. “India is not a country of risk capital at all,” Asokan said. Every pitch meeting turned into a classroom. “I spent eight years educating every investor in this part of the world,” she said. “They would come, spend time with us, learn how it worked, and this happened over and over for years.” Despite all this, they kept finding believers and mentors, investors who pushed them to dream even bigger.
“We should have died long ago,” Asokan said. Competitors collapsed every time the hype cycle shifted: dozens of US and Israeli startups chasing fashion AI, image recognition, or machine learning operations. MSD survived deep-learning, image-recognition booms, and then the ChatGPT wave, with Chandrasekaran re-architecting each time and Asokan carrying the story into rooms thousands of miles away.
Often she did not even leave Chennai. US department store Macy’s, Latin America-focused online marketplace MercadoLibre, the Tata Group: customers signed contracts with founders they had barely met in person.
Early days
Ambarish KC joined as Mad Street Den’s employee No. 1 in May 2013, before the company’s registration was completed. There were effectively three people writing all the code. He arrived knowing only the basics of Python. Chandrasekaran, by contrast, moved between Python, C++ and CUDA (Nvidia’s Compute Unified Device Architecture) with ease.
The first thing MSD shipped, in 2013–14, was not enterprise software but a mobile game called Blink: two players staring into a phone, their live video streams pushed to MSD’s backend, every frame processed for blinking, smiling and head movement fast enough to feel real-time. It was a minor engineering fever dream. Blink was only the first application. They pointed the same engine at self-driving-car analytics, at ad tech. “People were like, ‘Holy shit’,” Asokan remembered. “But they were like, ‘There’s no market for all this. This is 2013.’” So they kept rotating the engine until something stuck: retail.
Roughly 1,300 people passed through MSD’s doors over the years, about half of them women. The proof of what the place produced showed up in where these employees went next: some became early employees of OpenAI, others joined Apple’s AI teams.
- $50 million | The Series D that collapsed on signing day, December 2023.
- ~$60 million | Total capital raised by Mad Street Den, per public records.
- ~1,300 | People who passed through MSD over the years, about half of them women.
- 2,000–5,000 | Requests per second the 2015-era system processed, serving personalised results in 200–300 milliseconds.
- 2 weeks | The time it took Anand Chandrasekaran to build a vector search engine in 2015, five years before the industry had a name for the pattern.
Beating Palantir from Chennai
Asokan and Chandrasekaran’s ambition was always bigger than retail. They were not trying to be a feature vendor tucked into a supply chain. The story they told investors, advisers and themselves was infrastructure: a generalized AI operating system for enterprises, built out of India. A Palantir for the enterprise.
Some rooms leaned in. Others didn’t. In Bay Area meetings in 2017 and 2018, they were told by a Palantir backer that beating the US defense-focused AI firm from Chennai was unlikely without American “firepower” or access to government. They were not chasing defence dollars or myth-making. They wanted the enterprise version of that ambition: a horizontal platform that could sit inside any large organisation’s stack and quietly do the work.
Early on, Alfred Lin of Sequoia Capital had listened to their story and believed in what they were building. “You guys can do this,” he told them. But then he added: “Anything but retail.” No retail-tech company had crossed $100 million in revenue, he said. Start elsewhere, prove the platform, then take on the hardest vertical.
They didn’t listen. They couldn’t. In 2016, Indian e-commerce companies were paying them $100,000 per contract, serious money for a bootstrapped team in Chennai. By 2017 they were running catalogue automation and visual search at Macy’s. When you are winning deals that size, product-market fit feels obvious.
“Rookie mistake,” Asokan said. “You don’t understand how to process the wins you’re having.” The wall arrived with the pandemic. Macy’s nearly went bankrupt. Retail oscillated violently with every macroeconomic shift.
And yet the larger bet was held. By the early 2020s, Palantir was pushing from defence into the same enterprise markets Mad Street Den had served for years. In a run of bids, MSD’s sales team began coming back with a new line in their updates: we beat Palantir. These were not tiny pilots. In at least one instance, a US healthcare company chose MSD over Palantir. “The team started coming back and telling us, ‘Hey, we won a deal against Palantir,’” Asokan said. “That’s when it hit us—we actually had a chance at this.”
Closure, not collapse
The news of Mad Street Den’s end arrived in fragments: a whisper in founder WhatsApp groups, a business-press headline weeks later. Founders who had watched Asokan and Chandrasekaran teach the market for years called it premature. Why now, they asked, after Fortune 500 proofs, an AI platform built years before the boom, a handful of wins against Palantir?
What followed was a choice among imperfect options. What they went for, Asokan says, was global optimisation: a way for the people who had built the company to walk away with something, while some investors still saw a return.
Then she did something surprising: she put the acquisition offers on Slack (contours only, no names) and asked the leadership team to vote. “We sold to the company the team picked,” she said. “It’s been an incredibly rewarding journey to build something like MSD and wake up knowing the team that stayed till the end had a great payout.”
In March 2025, business news site Moneycontrol reported a cash-and-stock deal in which Chennai-based M2P Fintech would take on Mad Street Den’s intellectual property, contracts and employees. It pegged the deal at $10–15 million. People closer to the transaction described a far higher number, between $50 million and $100 million.
Even the most celebrated labs burn staggering sums merely to keep general-purpose AI models running, and early-stage AI startups raise hundreds of millions just to cover infrastructure. MSD had been processing images, documents and video in the cloud for nearly a decade before the current boom: billions of items through a generalised AI stack at a time when graphics processing units were scarcer and dearer, and the unit economics made almost no sense.
In India, the capital MSD had raised—about $60 million, per public records—looked like a large number. On the world map, it was not in the same league as the war chests its successors now assemble as a matter of course.
“Every time you raise investor money, you’ve started a clock. It’s a much shorter clock than a clock that you run on your own time and your own capital,” said Anandamoy Roychowdhary, a former Sequoia India partner and an early backer of Mad Street Den. “You think you have nine months, you think you have 12 months, but you actually have three.”
He added: “Honestly, this company had only one real problem: it was ahead of its time. Two years later, right now, it would have been a very nice generative AI company. It would have been a billion-dollar valuation.”
For Asokan and Chandrasekaran, the six months before the deal with M2P Fintech closed were the worst. Asokan lost the joy—not temporarily, but in the way her children could see it, which is the version that matters. Chandrasekaran was equally disturbed. “We are trapped in this situation,” he told her. “We have to get out.”
Both of them went under at roughly the same time, neither was available to hold the other up. There were panic attacks at 2 am. Then they would get up and do it again. For six months, Asokan looped the same Ben Horowitz podcast, ‘IPO from Hell’, letting someone else’s nightmare drown out the noise in her own head.
“Sometimes you just know when it’s time,” she said.
“What was the puzzle we were trying to solve?” Chandrasekaran said. “A full-stack, AI-based operating system for enterprise. We finished. We solved the puzzle.”
Chapter One
Asokan and Chandrasekaran do not describe Mad Street Den as the end of a journey. They call it Chapter One in a 50-year roadmap they have been sketching for decades.
“We got the timing wrong, there’s no denying that,” Asokan said. “But we got the AI transformation, the value, the delivery, the team and our vision for the generalized AI engine right.” There is more on the roadmap still: software, hardware, robots, maybe even space. “This is just the beginning.”
When Asokan was in hospital dealing with a health matter recently, Chandrasekaran sat in the room and coded for three days straight. They returned with more ideas and experiments than they had managed in weeks of ordinary working life. “We had days of quiet, peace, brainstorming about everything from spacetech to biotech without interruptions, just like back in the day (barring the surgery, of course),” she said. “It was so fun.”
Most evenings now, they walk for an hour, on the terrace, through new neighbourhoods. They do not make lists of shiny ideas. They make lists of traps to avoid: no go-to-market model that depends on a single internal champion; no architecture buried so deep the market cannot see the product; no investor table where the moral math fails in daylight.
Talk to them about the problem—any version of the problem, the architecture, the platform, the next chapter—and they light up. Twenty-five years in, that has not changed. Mornings still begin at 6:30 with coffee, milk and a board game before school. Those rhythms are what they carry forward.
Asokan has had this thought in her mind for a while: “Us walking away does not address the fact that we didn’t fight for it after a point. That we let some real believers down. And were OK with it because it was the hard call that had to be made. Is it not something that would be in everyone’s minds? That we are portraying walking away as a thing, almost a good thing.”
She went on. “I feel like what allows for this narrative to sit is a conversation about failure and success. And how India genuinely does not have a culture of encouraging failure. It’s treated shamefully. As opposed to the Valley, where that culture genuinely encourages risk and failure as the stepping stone to greatness.”
“I don’t want to sound like I know all the answers,” she added, “just a lot of questions that shape what and how I do what I do next.”
Pankaj Mishra is a journalist and author, and the founder of FactorDaily. A longer version will appear on FactorDaily later this year, followed by a live journalism session.
For more such in-depth reportage, read Mint Long Story