I got tired of keep optimizing prompt and rolling out changes in production. So I built a neurosymbolic language to help with that, where prompts as policies instead as hard rules.

https://github.com/canto-lang/canto-lang