Shenmin worked on the alibaba/rtp-llm repository, delivering foundational support for multimodal conversational processing by integrating the DeepSeekVLV2 model. He developed a new tokenizer class and enhanced the model architecture to process both image and text inputs, enabling richer interactions within enterprise chat applications. Using Python and leveraging deep learning techniques, Shenmin focused on natural language processing and multimodal processing with transformers. His work addressed the need for seamless integration of visual and textual data, laying the groundwork for future multimodal features. The depth of his contribution is reflected in the robust pipeline and architecture improvements he implemented.