Imagine building an App that simulates arguments in a court case between a prosecutor and the defense counsel with the judge listening and who finally makes a judgement on the case.
To keep it truly unbiased, we use independent Agentic frameworks and LLMs. The prosecution is built with agents using AutoGen and Claude, the defense uses CrewAI and Gemini Pro and the Judge Agent is built using GPT-4o and LangGraph.
The knowledge base is a common case file that is available to both sides. Based on the arguments, the agentic Judge decides on the final verdict. To push it a bit further, we could have 10 jury members (agents) who will also listen to the case offer their unanimous decision.
So, Is this possible in the current state of AI ?