See, Think, Learn: A Self-Taught Multimodal Reasoner