Abstract: The present disclosure involves methods, apparatus, and systems for customizing artificial intelligence (AI) models by performing automated, on-demand training. This can include, receiving, from an entity, a prompt for a generative AI model; updating a first interaction record with the prompt, the first interaction record comprising a history of interactions between the entity and the generative AI model; in response to determining that the updated first interaction record is greater than a predetermined size: loading the generative AI model into a training environment; loading one or more adapters in the training environment; training the one or more adapters in the training environment with the generative AI model based on the first interaction record; updating a second interaction record with the history of interactions in the first interaction record; clearing the first interaction record; and loading the trained one or more adapters with the generative AI model in an inference environment.