undefined | Better HN

0 pointsatarus2mo ago0 comments

Training is an overkill at this point imo. I have seen agents work quite well with a feedback loop, some tools and prompt optimisation. Are you doing fine-tuning on the models when you say training?

0 comments

rush869992mo ago

Nope - just use memory layer with model routing system.

https://github.com/rush86999/atom/blob/main/docs/EPISODIC_ME...

atarusOP2mo ago

Memory is usually slow and haven't seen many voice agents atleast leverage it. Are you building in text modality or audio as well?

j / k navigate · click thread line to collapse

0 comments

rush869992mo ago

Nope - just use memory layer with model routing system.

https://github.com/rush86999/atom/blob/main/docs/EPISODIC_ME...

atarusOP2mo ago

Memory is usually slow and haven't seen many voice agents atleast leverage it. Are you building in text modality or audio as well?

j / k navigate · click thread line to collapse