TechCrunch exclusively learned that Rohit Patel is leaving the company after eight years of developing AI for the meta. The former director of Meta Superintelligence Labs launched Ndoc, a startup pitching itself as an AI-native alternative to the Microsoft Office suite and Google Workspace.
Patel is the latest of several high-profile names to leave Meta, following the departures of chief international affairs officer Joel Kaplan, AI researcher Andrew Tulloch, and AI-for-Work product director Emily Dalton-Smith.
Patel designed Ndoc to be more than just an AI tool layered onto existing workflows. The goal is to replace fragmented ways of working with tools like Excel, Word, PowerPoint, Tableau, Replit, and “ChatGPT Copy and Paste” with one unified workspace that all colleagues can access, even if they haven’t signed up for Ndoc.
Ndoc combines three components into one layout. The (Stage) tab displays presentations of key findings, including slide decks, dashboards, simulations, and mini-apps. The Narrative tab shows analysis, numbered quotes, generated numbers, and generally the “why” behind the numbers. The (Model) tab stores the underlying dataset and Python script.
The design reflects Patel’s background. During his tenure at Meta, Patel focused on reinforcement learning (a type of machine learning in which AI agents learn how to make decisions and maximize rewards through trial and error) and assessing whether AI outputs can be trusted. Patel believes that if AI is building the spreadsheets and slide decks that companies use to make decisions, someone should be able to check how they work. For example, users can find the Python script used to create the document.
“Even if you don’t know anything about Python, you can ask[AI Chat]for clarification: ‘Hey, what does this script do? What line is my formula on?’ and you can validate that formula,” Patel said. “Nothing is hidden behind the curtain here.”
In practice, users can upload a dataset and ask Ndoc to perform an analysis and write a description of the results. You can also add document-specific context, such as audience, tone, and memory. “Memories are maintained for individuals, specific documents, teams, and the entire company,” Patel says.
Patel also said that Ndoc was built and fine-tuned using a combination of open source models. Because the platform was developed from the ground up, it costs significantly less to run than alternative platforms built on Frontier AI models, he says.
“I thought that’s what people would be excited about, but people don’t care about that,” he said, referring to early trial users. “What people are excited about is they like to see everything end-to-end in one place.”
Patel’s larger ambition is to reinvent the file itself. He said documents can be exported as a single .ndoc.html file that can be opened in any browser, and the company plans to open source the format, allowing other developers to build apps that work with it without being locked into Ndoc, as many developers have been locked away by Microsoft and Google for years.
His desire to open source the project was also part of the reason he decided to leave Meta and go it alone. At a large company like Meta, getting everyone to agree on an open source format will no doubt require input from “hundreds of people and their opinions on what the format should be.”
Patel wanted to act quickly. “Humans can only move at a human pace, but I can move at the same pace as an AI.”
Trying to be the platform where a company’s work is done is a big gamble for a founder with a three-person team and no funding, especially when other AI labs are looking to own the same space.
OpenAI, Anthropic, Meta, SpaceXAI, and Google have all announced enterprise AI tools that allow users to connect to Google Workspace, Microsoft 365, and other workplace apps, allowing AI models to work directly with these tools. And last month, OpenAI launched Space, a shared workspace within ChatGPT. This is a more direct competitor to these established tools.
Patel acknowledges that space can pose a threat, but says the institute’s thinking is largely wrong.
“Companies are basically attaching AI to a spreadsheet, and the AI will fill in your cells with data,” he said. “It’s like killing my soul. It’s like I just discovered electricity and decided the best use case was to put lights on the bulls that plow my soil. No, buy a tractor.”
But there’s a reason the AI Institute sought to deploy its models where people are already working. It’s hard to change your habits or replace your entire workflow. Patel says he understands that, which is why he built the connector to allow customers to import existing data from the tools they already use.
“My criteria was not to release anything I didn’t want to use,” he said. “I use highly optimized AI workflows for everything I do. If you start using this tool and keep going back to your own workflow, you haven’t built a good enough tool… There’s a graveyard of tools that don’t meet that standard.”
Patel has similar hurdles in place to gain customer attention before attempting to raise capital. “We didn’t want to take the money until we had proof that people liked the product,” he says.
As of Tuesday, Ndoc is running an enterprise pilot, opening an alpha version to a limited number of customers on a waiting list. Sign up is free, but customers pay usage fees and token fees. He plans to release a beta version in the coming weeks, with a full release planned for next year.
For Patel, acting as quickly as possible is more important than having the perfect product the first time.
“The cost of execution is very low… so do it with a suboptimal plan and worry about it later,” he said. “By the time you have your best plan, someone else will execute it.”
If you make a purchase through links in our articles, we may earn a small commission. This does not affect editorial independence.
