Reducing API costs is a massive priority for teams right now. Are you using a smaller model like Llama 3 for the local filtering layer?