Jesse Silver: recommendation

Transcript context

…So I wanted to get more technical. We've started talking a little bit about your state machine, the models that you use. Could you just describe your tech stack in whatever way you think is interesting for engineers? What big choices you made? What did you evaluate and didn't go with? Anything like that? At the start, we had a very simple product that had a limited amount of language bottle generation. And based on this, we started using sort of low code prototyping tools to get a workflow that worked for a limited number of creators or a limited number of cases. But I think one of the biggest challenges that we faced is just the raw number of times where we've put the product on an account and it just sucks. And we have to figure out why. And the creator will say things like, I can't believe you sold something for $11, 13 makes so much more sense. And we're like, oh, like there's a whole part of the world that doesn't exist. And so in the start, a low code prototyping platform was very helpful in trying to understand what a sort of complete model would look like. And then it got sort of overburdened. And we decided to move to DSPy. And we wanted to take advantage of the ability to optimize things on the fly, have a more elegant representation of the workflow, keep things in Python, and also easier way of fine tuning models on the go. Yeah, and I think the other piece that's important is the way that we evaluate things. And I can talk about that as well, if that's of interest. Yeah, you said you had your own eval framework. Probably that's something that we should dive into. I imagine when you're model shopping as well, I'm interested in basically how do you do evals?…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.