For me it's using 1 Jev call to ask 5 questions, instead of what I had before.

What I had before was basically:

- 1 classifier question

- if ok, do some heavy computation

- after heavy computation, another classifier

So now what I do is:

- Jev with the 1st and 3rd classifier

- if ok, heavy computation

I can classify many things at once, and it's cheaper and faster than 1 for sure and 1 maybe classification that I had before.

^ Same here. so much faster and cheaper than using an LLM to return structured data