Super cool article. I could see this tech getting mass adoption once the latency is low enough...

Yep, latency is the biggest challenge we're solving. In-housing core infra is the best thing we can do in the short term (although expensive).

Other option is that type of hacky acknowledgement we described - still not sure if we're going to do this or not.

Excited to be sharing our work with the HN community; here to answer questions and talk through the models we use, how we're building 'conversational intelligence,' and all the other aspects of AI calling.