Former Meta AI director Rohit Patel launches Ndoc to challenge Microsoft Office and Google Workspace

Rohit Patel has left Meta, where he spent eight years developing AI, to become founder and CEO of Ndoc (ndoc.ai). The startup positions itself as an AI-native replacement for Microsoft Office and Google Workspace. TechCrunch reported the move on 27 October 2026.

At Meta, Patel was a director at Meta Superintelligence Labs. His work centred on reinforcement learning, a training method in which a model is rewarded for outputs that score well, and on evaluating whether AI output is trustworthy. Both bear on how a generative product handles factual claims. A Forbes Technology Council profile lists him as Founder and CEO of ndoc.ai.

Ndoc is organised into three tabs. The Stage tab produces presentations such as slide decks, dashboards, simulations, and mini apps. The Narrative tab holds analyses with numbered citations and generated figures. The Model tab houses the underlying datasets and Python scripts, the code that performs the calculations. According to the same report, a user can upload a dataset, ask the system to run an analysis, and have it write a narrative about the findings. The product is built and fine-tuned, meaning further trained for its specific job, using a mixture of open-source models, which are AI models whose code and weights are publicly released.

The output format is the detail most likely to draw engineers' attention. Ndoc documents export as a single ".ndoc.html" file that opens in any browser. Patel plans to open source the format. That is a stated intention, not a released specification.

Patel's earlier public work sits in the background. In October 2024 he announced Emu Video and Emu Edit as new generative AI research milestones in a LinkedIn post. Those were research releases in image and video generation, a different product lane from document and analysis tooling.

The broader context here is a crowded field. Office productivity has been dominated for decades by suites that bundle documents, spreadsheets, and presentations into server-hosted, subscription-based services. Ndoc's pitch relies on a different architecture: generation and analysis are built into the document itself, and the output travels as a self-contained HTML file rather than a proprietary container that depends on a hosted application. Whether that portability holds up in practice depends on how the format is specified and whether the open-source release arrives as described.

In my view, the questions that matter most for Ndoc are less about the feature list than about verification. A tool that writes a narrative from a dataset and attaches numbered citations and generated figures is only as useful as the trust a reader can place in each figure. Patel's Meta work centred on that evaluation problem, and the Narrative and Model tabs suggest the product is designed to expose its reasoning and its code. How well that traceability holds under real analytical workloads is the test.

For practitioners, the near-term signals are concrete. Watch whether the .ndoc.html specification is published, which open-source models are named, and how the Model tab's Python output handles reproducibility when a user reruns an analysis. If the format becomes genuinely portable and the analysis is inspectable, a self-contained, browser-readable document with its own provenance could be a useful addition to the office toolkit. If not, Ndoc will join a long list of AI-native productivity launches that impressed in demonstration and struggled in daily use.