The first clue is a shopping problem. A person sees a pair of shoes in a photograph and wants to find something like them. Search engines can understand words; the photograph offers shape, color and texture. In the middle of the 2000s, Munjal Shah built Like.com around that gap. The company let shoppers search by visual resemblance, turning computer vision into something that could be tried in an ordinary browser. Google bought it in 2010. The product belonged to a moment before every phone camera seemed to come with a search bar.
Now the problem is a conversation. Shah’s current company, Hippocratic AI, builds voice agents that speak with people on behalf of healthcare organizations. It is a very different market with different stakes. Yet the question sounds familiar: can complicated AI be shaped into an interface people will actually use? Shah has spent much of his career answering that question, first with forms and listings, then images, and now speech.
His résumé can be read as four companies and a collection of exits, jobs and investments. It reads more clearly as a history of interfaces. Each new venture takes a capability that engineers recognize and asks what it would look like in the hands, or ears, of everyone else.
A thesis before the fashion
At the University of California, San Diego, where Shah graduated in 1995, his senior thesis used neural networks to predict how molecular structures would interact. He began his career at Agouron Pharmaceuticals using neural networks in research. He went on to earn a master’s degree in computer science at Stanford, with an emphasis on artificial intelligence. These details matter because the current enthusiasm for AI can make every founder sound newly converted. Shah was working with the underlying ideas long before they became a standard line in a pitch.
The first company he built did not look like an AI company at all. Andale, founded in 1999, made software for small businesses selling through online marketplaces. The internet was filling with people who could list goods but still needed help managing the details of selling them. Andale addressed the back office of that new trade. It was a practical founding lesson: a new platform creates a second market of people trying to work within it.
A founder can spend a career chasing novelty. Shah’s early moves suggest he was more interested in friction. For an online merchant, the friction was managing a business across marketplaces. For a shopper, it was the awkwardness of describing an object that was easier to recognize than to name. The technology changed; the instinct to smooth an everyday task did not.
The search box learns to look
Like.com was an unusually literal answer to a common shopping sentence: “I want one like that.” Its software used computer vision to compare the visible qualities of products. Former colleague Vineet Buch described Shah as a chess player when it came to running a company, someone who thought several problems ahead. Former colleagues also remembered a founder intensely focused on execution, with a taste for deadpan, self-deprecating jokes. One colleague’s verdict was neatly phrased: “You don’t have to be serious to be taken seriously.”
It is a useful description of the gap between a founder’s pitch and the labor underneath it. A visual-search demonstration is easy to understand; a business that makes it work for actual shoppers is harder. Like.com had to be useful when the photograph was messy, the product catalogue incomplete and the customer impatient. By the time Google acquired it in 2010, the idea had moved from a clever visual trick to a credible shopping tool.
Shah stayed at Google for a period, working in Shopping product management. The move gave him a view from inside the giant that had bought his startup. It also put him close to a larger distribution system. A founder who once had to persuade people to try a new search behavior was now working where millions of shopping searches already happened.
“You don’t have to be serious to be taken seriously.”Gaurav Suri, former colleague, on Shah’s dry humor
A detour with a long shadow
In 2013, Shah founded Health IQ. The company initially used data about lifestyle and behavior in an insurance business and later expanded into matching people with Medicare plans. It was a move away from product search and toward a tightly regulated service market. Its story was more complicated than a smooth Silicon Valley exit: Health IQ filed for bankruptcy in 2023. The experience belongs in the account because it shows that Shah’s career has not been a procession of clean victories. He spent a decade learning what happens when software meets institutions, rules and expensive human processes.
During those years he was also active as an investor. He has backed more than 42 startups and 23 early stage venture funds. He supports entrepreneurial education at his alma mater through the Shah Fellows program. The founder who builds his own companies has also kept a seat near other people’s early attempts, where a product is still an argument and the market has yet to answer.
That breadth matters to the latest chapter. Hippocratic AI did not emerge from someone discovering a chatbot over a weekend. Shah had worked with early machine learning, built consumer AI, seen Google’s scale, and run a business in a highly structured industry. Each part of that history supplied a different warning about how easily a promising demo can outrun the work of making it dependable.
A voice, with boundaries
Shah co-founded Hippocratic AI in 2023 with a group that included AI researchers and healthcare professionals. The company’s focus is voice agents for nondiagnostic tasks: conversations and follow-up work that organizations struggle to do at scale. The choice of voice is important. A web form waits for someone to arrive. A phone conversation can go to them. The interface is familiar enough to require no instruction, which also makes it easy to overestimate.
Shah has described an unusually long pause before the first deployment. In a 2026 public post, he said the company spent 18 months testing before an agent spoke with a patient. He outlined boundaries around what the agents can do and described real-time oversight by independently trained models. The details are company statements, and they offer a window into his chosen operating logic: define the job narrowly, test it repeatedly and treat the voice itself as a consequential product choice.
The name Hippocratic AI brings a heavy promise into a young company. A voice can sound warm while being wrong; a useful conversation has to be judged by more than tone. Shah and his team have put safety language at the center of the company’s identity, including the phrase “do no harm.” The work of making that phrase concrete belongs in evaluation, limits and escalation, where the public rarely sees a product’s labor.
There is a human tension in the idea. Technology companies love to describe scale as though it were a virtue by itself. Shah’s account asks whether an interaction can remain appropriately bounded as the number of interactions rises. The answer is still being tested in public, through customers, clinicians and the people on the other end of the line.
Hippocratic AI’s reported figures as of November 2025.
The size of the bet
The company’s expansion has been fast by its own account. In November 2025, Hippocratic AI announced a $126 million Series C financing at a $3.5 billion valuation. It reported more than 50 large client organizations in six countries and more than 115 million interactions during the 15 months after commercialization. Those are company reported numbers, useful for understanding the ambition and scale, though they are not a substitute for independent measures of what each conversation accomplished.
The investors include General Catalyst, Andreessen Horowitz, CapitalG, Kleiner Perkins and others. General Catalyst’s Hemant Taneja has described himself as a co-founder of the company. The partnership puts Shah inside a familiar Silicon Valley arrangement: substantial capital, a large institutional market and a timetable that pushes a young company to build while it learns. Yet the product is heard rather than seen. There is no shiny new object to admire, only a voice and the system behind it.
In August 2026, the company announced a move from individual agents toward “orchestrators” that coordinate specialized agents around broader organizational goals. The idea is a natural extension of Shah’s old interest in workflow. One task can be automated; the more difficult problem is deciding which task should happen next, who should do it and when a person needs to take over. His current language of “abundance” is an aspiration for more capacity. The actual achievement will be measured in the small, repeated decisions of those coordinated systems.
Shah has taken that argument beyond product announcements. In September 2026 he joined a CSIS discussion about AI startups and regulation. The appearance is a reminder that this venture sits where technical design, institutional trust and public rules meet. His earlier companies had their own forms of friction. Few required the founder to explain boundaries quite so often.
“Safety is a decision a CEO makes, and then proves with actions that match.”Munjal Shah, public statement in 2026
The familiar question
There is an appealing neatness in saying that Shah has spent thirty years making machines more human. That would simplify the record too much. Andale was about commerce, Like.com about vision, Health IQ about insurance, Hippocratic AI about voice. The through-line is more specific: he repeatedly finds a new technical capability at the edge of practical use and builds a business around the last difficult step between demonstration and daily work.
His career also resists the standard portrait of the solitary visionary. There are co-founders, investors, former colleagues, researchers and the organizations that choose whether to use a product. The chess comparison captures part of his style, but a board has many pieces. Shah’s ventures have depended on teams turning an insight into a system that survives contact with ordinary people.
At Like.com, the test was whether a shopper could find the thing she had seen. At Hippocratic AI, the test is whether a conversation can be useful, bounded and worth repeating at enormous scale. The newer problem is less photogenic, and the standards are higher. Shah has reached the point in a founder’s story where the most interesting line is no longer “what can the machine do?” It is the quieter line that follows: who decides when it should speak?