Building an open source multi-modal AI assistant