Amazon Alexa

Every Alexa request passes through a layer that decides which skill or domain gets to answer. I owned it, and with it a dependency on every other Alexa team and about ten thousand external developers. The work that shipped: skill links so a customer could talk to two skills at once, passthrough so other voice agents could run inside Alexa, and latency work that gave everyone else more room. It is where the research questions I now work on came from.