Meta director Rohit Patel leaves to build Ndoc, an AI alternative to Office

Rohit Patel has left Meta after eight years to build Ndoc, an AI workspace pitched as a replacement for Microsoft Office and Google Workspace. The former Meta Superintelligence Labs director’s startup is running an enterprise pilot and opening a limited alpha, betting that workers want their analysis, explanations and presentations together—not another assistant beside existing tools.

In an interview with TechCrunch’s Rebecca Bellan published October 27, Patel described a product designed to replace the back-and-forth between spreadsheets, documents, slide decks, coding tools and copied chatbot answers. His ambition is not simply to generate each output faster. It is to put the work behind those outputs in the same place, accessible to colleagues even if they have not signed up for Ndoc.

One workspace, three views of the work

Ndoc’s layout separates the finished presentation from its explanation and supporting material without sending users into different applications. A user can upload a dataset, ask for an analysis and have the system write a narrative about the findings. The workspace then organizes that work into three tabs:

• Stage holds presentations of the findings, including slide decks, dashboards, simulations and mini apps.
• Narrative contains the analysis and explanation, with numbered citations and generated figures.
• Model houses the underlying datasets and Python scripts—the code used in the work.

Users can also supply context for a particular document, such as its audience and tone. Patel says Ndoc maintains memory at the individual, document, team and company levels. That is part of his pitch for a shared working environment rather than a series of isolated requests to a chatbot.

Checking the formula, not just reading the answer

At Meta, Patel worked on reinforcement learning—training AI through rewards and trial and error—and evaluating whether AI output could be trusted. That background informs Ndoc’s emphasis on exposing the machinery behind a document. Users can find the Python scripts used to create it, rather than receiving only a polished result.

Patel says people who do not know Python can ask the AI chat to explain what a script does and identify the line containing a formula. His argument is that users should be able to inspect the work behind spreadsheets and presentations used for business decisions. The proposed check still involves the user: Ndoc makes the script available and offers an explanation, rather than removing the need to verify it.

He also says Ndoc uses a mixture of open-source models that it has fine-tuned, and claims that building the platform from scratch makes it substantially cheaper to run than alternatives using frontier models. But Patel told TechCrunch that early users were more interested in seeing the entire workflow together than in the underlying model economics.

A replacement that still needs a way in

Patel criticizes the approach of attaching AI to existing spreadsheets, but acknowledges the difficulty of changing established workflows. Ndoc includes connectors for importing data from tools customers already use. His larger ambition is to change the file itself: documents can be exported as a single .ndoc.html file that opens in a browser. He plans to open-source that format so other developers can build compatible applications; that remains a plan, not a completed release.

For now, the venture has three people and has not raised funding. Signup is free, but customers pay for usage and tokens, the units used to measure AI processing. Alpha access is limited to selected waitlisted customers. Patel hopes to release a beta in the coming weeks and plans a full launch in 2027.