karpathy/llm-council ? reverse-engineered prompt

Build me a simple, local web app that lets me ask questions to a "council" of different LLMs. When I submit a prompt, it should first send it to several models at once—like GPT, Claude, and Gemini. I want to be able to see all their individual responses, maybe in a tabbed view.

Then, the cool part: each model should anonymously review and rank the others' answers for accuracy. Finally, a designated "Chairman" model should take all the initial responses and the reviews to compile one definitive final answer for me.

Let's use Python with FastAPI for the backend and React with Vite for the frontend. The app will need to use the OpenRouter API to access all the models, so make it easy for me to add my API key using a .env file. I'll also want a simple config file in the backend where I can easily change the list of council models and pick the chairman model.