in a nutshell an LLM interprets texts/images/... by converting inputs to vectors. It compares these vectors to others in a vector database. It reasons about which words are associated with eachother in this way. The algorithm used to convert a text to a vector is the model. The resulting vector is the embedding. Some models are free and can be found on sites like HuggingFace. The lingua franca right now seems to be Python, models often are provided as pytorch files so embeddings have to be calculated in that language. Not sure what models work with Umbraco though