A hypothesis on how to generate better results with LLM. You force one model to act as several personalities, and then to evaluate its own answers.
Dart language is used for both frontend and backend, making use of google_generative_ai package. All the magic lies in well-prepared prompts.
Live demo (in Russian): https://peaashmeter.dev/schizodem/