Former Meta AI Director Rohit Patel Departs to Launch Ndoc, an AI-Native Challenger to Microsoft Office and Google Workspace

After spending eight years developing advanced artificial intelligence systems for Meta, Rohit Patel is stepping away from the tech giant to embark on an ambitious entrepreneurial venture. Patel, who formerly served as a director at Meta Superintelligence Labs, is launching Ndoc, a newly unveiled startup designed from the ground up to serve as an AI-native replacement for entrenched enterprise ecosystems like the Microsoft Office suite and Google Workspace.

Patel’s departure marks the latest in a wave of high-profile exits from Meta’s ranks. Over recent months, the company has seen the departures of several key leaders, including Chief Global Affairs Officer Joel Kaplan, prominent AI researcher Andrew Tulloch, and Emily Dalton Smith, the head of product for Meta’s AI-for-work transformation initiatives. These transitions highlight a broader industry shift as top-tier talent leaves major laboratories to pursue independent visions in the rapidly evolving enterprise AI landscape.

Unlike many current enterprise tools that merely overlay AI features onto traditional workflows, Patel designed Ndoc to fundamentally reimagine how teams handle data, documents, and collaboration. The core premise of the startup is to eliminate the fragmented experience of jumping across multiple siloed applications—such as Excel, Word, PowerPoint, Tableau, Replit, and traditional "ChatGPT copy-paste" routines—and replace them with a single, unified workspace. Notably, Patel engineered the platform so that colleagues can access and view these collaborative spaces even if they do not maintain active Ndoc accounts themselves.

The platform’s interface is structured around three distinct components integrated into a single layout. The Stage tab functions as a home for presentations containing key findings, including slide decks, live dashboards, simulations, and mini-applications. The Narrative tab houses deep-dive analyses, numbered citations, automatically generated figures, and the contextual "why" behind numerical data. Finally, the Model tab provides complete transparency by storing the underlying datasets and Python scripts that power the analyses.

This design is a direct reflection of Patel’s professional background at Meta, where his work heavily centered on reinforcement learning—a machine learning methodology where AI agents learn to make decisions through trial and error to maximize rewards—and the rigorous evaluation of AI trustworthiness. Patel holds the firm conviction that if artificial intelligence is going to author the spreadsheets and slide decks that corporations rely upon for major decision-making, users must be able to independently verify the AI’s underlying work. By allowing users to inspect the Python scripts used to generate corporate documents, Ndoc aims to remove the traditional "black box" nature of generative AI.

Addressing the transparency of the system, Patel explained that users do not need to be programming experts to audit the software. Even individuals with no prior knowledge of Python can interact with the system’s chat interface to ask specific operational questions, such as identifying what a script does or locating a specific formula within a line of code. This open architectural philosophy ensures that nothing remains hidden behind the curtain.

In daily operation, Ndoc allows users to upload raw datasets and prompt the platform to execute complex analyses accompanied by comprehensive written narratives explaining the discoveries. Furthermore, users can enrich the output by injecting document-specific context, including parameters for the intended audience, desired tone, and historical memory. The platform is designed to maintain institutional memory across individuals, specific documents, teams, and the broader corporate enterprise.

Under the hood, Ndoc is constructed and fine-tuned using a strategic mixture of open-source models. Because Patel developed the platform from scratch rather than relying exclusively on heavy proprietary architectures, he notes that it costs considerably less to run than alternatives built entirely on standard frontier AI models. Initially, Patel anticipated that these underlying cost efficiencies would be the primary selling point for prospective users. However, feedback from early trial users revealed a different priority: enterprises were far more captivated by the seamless, end-to-end consolidation of their daily workflows into a single location.

Patel’s broader vision extends to reinventing the fundamental structure of the digital file itself. He has engineered Ndoc so that complex documents can be exported as a single file format carrying the extension “.ndoc.html.” These files are designed to open natively within any standard web browser. Patel intends to open-source this specific file format, allowing independent developers to build external applications that integrate with it, thereby freeing the developer ecosystem from the traditional proprietary lock-in enforced by tech giants like Microsoft and Google for decades.

This desire to champion open-source standards was a significant catalyst behind Patel’s decision to leave Meta and launch his venture independently. Operating inside a massive organization like Meta often requires extensive consensus-building, where gaining approval for a standardized file format would necessitate navigating the opinions and feedback of hundreds of stakeholders. Opting for independence allows Patel to operate at a vastly accelerated tempo.

Attempting to establish a new foundational platform for enterprise work is a monumental gamble for a founder operating with a lean team of three and zero outside funding, particularly given the fierce competition from established AI powerhouses targeting the exact same domain. Industry leaders including OpenAI, Anthropic, Meta, SpaceXAI, and Google have all rolled out enterprise-focused AI tools designed to integrate directly with Google Workspace, Microsoft 365, and other legacy workplace suites. Furthermore, OpenAI recently introduced Space, a shared workspace integrated directly into ChatGPT that poses a more direct challenge to traditional office software.

While Patel acknowledges that products like OpenAI’s Space represent a formidable competitive threat, he maintains that major AI laboratories are largely approaching the problem from the wrong conceptual angle. In his view, existing tech labs are merely attaching generative tools to traditional spreadsheets so that AI can automatically populate cells, a practice he views as a misallocation of technological potential. He draws a vivid parallel, suggesting that using powerful new AI simply to fill cells in legacy spreadsheets is akin to discovering electricity and deciding the best application is to strap lightbulbs to the beasts of burden tilling agricultural fields, rather than building a tractor.

At the same time, Patel acknowledges the massive friction inherent in trying to change entrenched human habits and replace decades-old corporate workflows. To ease this transition, he built connectivity features into Ndoc that allow customers to effortlessly import existing data from the tools they already use daily. His guiding principle throughout development was uncompromising: he refused to release any feature or workflow that he did not personally love using. Having accumulated a personal graveyard of workplace tools that failed to meet his rigorous standards, Patel ensured Ndoc cleared his own productivity bar before bringing it to market.

In alignment with his philosophy on product validation, Patel deliberately chose to launch the business without seeking venture capital funding. He wanted to secure authentic proof that real users genuinely loved the product before inviting outside capital into the firm.

As operations move forward, Ndoc has initiated its first enterprise pilot alongside an alpha program opened to a carefully limited number of customers from its waitlist. While signing up for the platform is free, customers operate on a usage-based pricing model tied to token consumption. Patel anticipates rolling out an expanded beta version in the coming weeks, followed by a full commercial launch planned for the following year. For Patel, maintaining relentless momentum and execution speed remains vastly superior to waiting for a flawless initial product rollout. In his view, execution is exceptionally affordable, meaning founders should execute on a suboptimal plan immediately rather than lose momentum waiting for perfection, because by the time a perfect plan is finalized, someone else will have already executed the idea.