We use public datasets, but Umme's core is not based on LLMs. Similar to Google AlphaGo, our approach relies neither on LLMs nor neural nets.