Meta AI Director Rohit Patel Launches Ndoc, an AI‑Native Workspace to Replace Office Suites

After eight years shaping reinforcement‑learning research at Meta, Rohit Patel has stepped down to build Ndoc, a startup positioned as an AI‑native alternative to the traditional office suite. The platform merges three core views — Stage for presentations and dashboards, Narrative for cited analyses and generated figures, and Model for the underlying datasets and Python scripts — into one collaborative workspace that even non‑registered teammates can access.

Patel’s background in evaluating AI trustworthiness drives a design that makes every calculation auditable. Users can inspect the exact Python code behind a chart, ask the embedded chat to explain any line, and verify formulas without hidden logic. The system is fine‑tuned on a blend of open‑source models, which Patel says keeps inference costs well below those of frontier‑model‑based competitors.

Beyond the product, Patel is rethinking the file itself. Documents export as a single .ndoc.html file that opens in any browser, and he plans to open‑source the format so other developers can build compatible tools without vendor lock‑in. This move mirrors his desire to move at “AI speed” rather than the slower consensus cycles of a large corporation.

Why it matters for GPU / AI infrastructure

Ndoc’s reliance on curated open‑source models means inference workloads can run on modest GPU clusters, lowering the barrier for enterprises that lack massive compute budgets. A unified workspace that stores code, data, and narrative together also creates a natural pipeline for continuous model retraining, driving steady demand for scalable GPU resources. Moreover, an open file format encourages a broader ecosystem of AI‑enabled productivity apps, expanding the total addressable market for AI‑accelerated cloud services.