Distildoc: Deep Reinforcement Learning for Token-Efficient Multimodal Document Question Answering. Int Res J Adv Engg Hub. 2026;4(05):3775-3782. doi:10.47392/IRJAEH.2026.0496