Amazon Alexa

Every Alexa request goes through a layer that decides which skill or domain gets to answer it. I owned that layer - and with it, a dependency on every other Alexa team and ~10,000 external developers.

What shipped:

1. Skill links, so a customer could talk to two skills at once.
2. Passthrough, so other voice agents could run inside Alexa.
3. Latency work that gave everyone else more room.

This is where the research questions I work on now came from.