Mukund Sridhar

Mukund Sridhar helped develop Gemini Deep Research, an AI research assistant that browses the web and assembles its findings into detailed reports. In February 2025, he was its tech lead, working alongside lead product manager Aarush Selvan. Their collaboration combined a concrete research problem—gathering scattered information and comparing options—with a system that could produce a report for the user.[1][2]

Sridhar and Selvan developed the idea around a challenge given to Selvan: finding the right summer camp for children required comparing prices, availability and schedules across many websites. Selvan recalled that, after a few weeks of prototyping with his team, Sridhar brought him into a room to demonstrate the system. Sridhar asked it to find and compare soccer and art summer camps in the New York area. He reported that it found details for more than 20 camps: “So we thought, OK, we have something really cool here.”[2]

Sridhar described Deep Research as helping people “go from zero to understanding a subject deeply”. His own use of reports included further editing and collaboration: “If I really like the report, I want to edit it myself and share it with people for their edits and comments.” Exporting a report to Google Docs carried its citations into a works cited section, supporting that workflow.[2]

In the February 18, 2025 Latent Space interview, Sridhar identified personalization as a direction he wanted research agents to pursue. A report for a 15-year-old high-school student, he argued, should differ from one for a PhD or postdoctoral researcher. Beyond responding to explicit prompts, he wanted an agent to understand what a person already knew and where they were in their learning journey.[1]

Sridhar also wanted multimodal input and output, including interactive reports with charts, maps and images. He proposed generative UI as a way to customize the structure and presentation of a report for its purpose: an academic paper and a learning program for a child would call for different forms. These were interests he expressed for future research agents in February 2025.[1]

Another priority was extending research beyond publicly accessible websites. Sridhar argued that agents would reach “the limits of the open web” and that users should be able to supplement research with their own documents, corpora and subscription content. He considered this especially relevant to specialized industries, where information people cared about might sit within those collections.[1]

Before Deep Research, Sridhar and Selvan had worked together on Gemini extensions. In the February 2025 interview, Selvan described extensions as the first feature they both worked on together on the team. That earlier work let Gemini fetch content from other Google services to inform its answers, providing context for their experience connecting the assistant to information outside the chat itself.[1]