The fastest way to turn your taste and expertise into automation. Arbiters are language models that make a decision, explain why, and return a (reliable) score for how sure they are -- measured from inside the model. Confident decisions are accepted, unsure decisions go to a human, then their fixes update the instructions automatically. Over time, arbiters align to your judgment and decision rules. Use them to score, classify, route, extract, and moderate -- the way your team would by hand.

DSPy is a framework for programming LLMs instead of prompting them manually. While the philosophy of Declarative AI Programming makes experimentation with LLMs a lot more methodical and composable, there's still a high barrier to create and optimize working DSPy programs (heavy DataOps and evaluation mindset). Modaic is a developer collaboration platform for the new primitives of AI software. An open ecosystem where researchers and devs can version, package, and collaborate on DSPy programs!