Bringing AI across Microsoft 365 before the category was obvious.

Speech and language were becoming product surfaces, not back-end utilities.

Neural nets were making translation, transcription, captions, accessibility, and usability feel like one product shift: software could finally understand more of the human context.

This was AI before the category became obvious to everyone. It had to work inside the software people already used, with no prompt box, no AI theater, and no explanation required.

Live captions and translated subtitles in PowerPoint mattered for deaf and hard-of-hearing audiences, international presentations, classrooms, conferences, and anyone trying to follow along across language or hearing barriers.

The model was only part of the work. Real-time, low-latency, jargon-aware language at global scale was hard. The experience was the hard part: useful, reliable, accessible, and understandable.

I founded and led the AI team inside Microsoft 365 for five years, across speech, language, translation, transcription, accessibility, usability, and some vision-adjacent work.

I helped bring AI into real product surfaces across Outlook, Word, PowerPoint, Teams-related work, and broader Microsoft 365 experiences.

The work included Translator, transcription, captions, subtitles, writing assistance capabilities, and prototypes that pushed where the suite could go next.

The products shipped to 100M+ users in 100+ markets and were used daily.

Scale makes vague thinking impossible. The product has to work for people who do not know or care how advanced it is.

Accessibility is not a side quest. It is one of the clearest ways to understand whether the interface is actually doing its job.

That work is the foundation for how I think now: AI product judgment, interface depth, accessibility, global users, real constraints, and shipped systems.

Then I left to build the next layer myself.