Bongard-mini

Bongard-mini is an open-source AI model implementing machine intuition and judgment using the T5 encoder-decoder architecture. It reads text, structured data, and images to answer complex questions by providing probability distributions over candidates. It supports zero-shot classification, semantic understanding, and action outcome prediction, suitable for tasks like customer conversation analysis, invoice reconciliation, and game move selection. The product includes Python and CLI interfaces, local serving capabilities, and a live demo on Hugging Face.