Amit Desai

Voice AI for assistants, devices, robots

I like to create magical voice interfaces for new devices, surfaces, and modalities.

I believe the best voice interface marries (1) a deep intuition for spoken human conversation — psycholinguistics, turn-taking mechanics, interactive dialog — with (2) a technical fluency in emerging AI capabilities (and limitations). Both are needed to solve voice challenges for each new surface: robots, multimodality, wearables, phone calls, earpods, in-car, hands-free ambient, and more.

I'm also interested in "interaction intelligence": optimizing a voice system's actions to minimize a user cost function. Stated simply: should the voice system act, confirm, or reject?

About me

I'm passionate about all things voice :) Currently I lead voice & multimodal AI at Roku. Previously in voice UX leadership at Alexa. I founded Voxify, a voice startup for customer support IVR agents (for United Airlines and others). I built a voice-assistant mobile app (Top 5 in iOS App Store). MIT, B.Sc. computer science. Background in writing and illustration.

Contact

Ideas & work

Selected concepts and products.

Presented at AI Engineer World's Fair 2026.

An under-invested part of voice AI is behavior under uncertainty: after interpreting a command, should the assistant act, confirm, or stop? I propose OUCH (Outcome User Cost Heuristic) as a decision policy to choose behavior that minimizes expected user cost — balancing wrong actions and redundant confirmation.

Read more

Lamp demo