Beijing, China
I recently joined the Qwen Application Foundation Model Team as a researcher, working on training and optimizing LLMs for Qwen App. My work in post-training bridges fundamental research and tangible gains for the Qwen App. Beyond improving application-level performance, I devote a substantial amount of effort to exploring and advancing the core capabilities of LLMs. I am particularly interested in pushing the boundaries of model training, optimization, and reasoning behaviors, with the goal of producing work that leads to more fundamental and impactful advances in large-scale models.
I am deeply engaged in the advancement of Datanet, a cutting-edge data federation system designed to seamlessly integrate a variety of data sources into a unified and efficient framework. My goal is to evolve this system into a data fabric architecture, enhancing its capabilities to not only connect data but also to process and analyze it intelligently, thereby improving data accessibility, interoperability, and overall smart data management across multiple platforms.
Driving a joint research effort with Stanford and Canlah to evaluate how cutting-edge large language models can transform application modeling. I lead the exploration of LLM-powered semantic understanding, system design automation, and intelligent workflow generation, helping establish new methodologies that bridge academic research with real-world software engineering.
As a summer intern in the Microsoft One Person Entrepreneur (OPE) program, I focused on exploring AI startup projects by harnessing the powerful capabilities of large language models (LLMs) to identify innovative use cases and unlock new application scenarios. I prototyped solutions that could significantly enhance user experiences and drive engagement through advanced AI integration. Additionally, I contributed to the development of generative recommendation systems, leveraging LLMs to improve the accuracy and relevance of recommendations. This experience allowed me to combine entrepreneurial initiative with technical expertise, while collaborating with cross-functional teams to bring cutting-edge AI solutions to life.
As a Large Language Model (LLM) researcher at Microsoft Research Asia, I am actively engaged in exploring the diverse applications of LLMs within the context of Excel. My role involves conducting cutting-edge experiments to harness the power of LLMs for enhancing productivity and automation in spreadsheets. I am dedicated to pushing the boundaries of what's possible by integrating advanced AI capabilities into Excel, aiming to transform the way users interact with data and perform complex analyses. Through my work, I am committed to uncovering innovative solutions that can streamline workflows and empower users to make data-driven decisions more effectively.
As a Software Engineer at Microsoft AI's Edge Machine Learning Team, I have made significant contributions to the full-stack development of the Edge Copilot project. My work has focused on creating an AI-powered browsing and search engine within Microsoft Edge and Bing, which enhances user experience by using LLM.
I was deeply engaged in pioneering research on Large Language Model (LLM) agents specifically tailored for the e-commerce domain. My work focused on developing a sophisticated conversational proxy that leverages the transformative potential of Dify, a platform designed to customize and enhance agent capabilities. By utilizing Dify, I ensured the seamless integration and optimal performance of our LLM agents within the e-commerce ecosystem. Additionally, I actively learned and applied advanced cloud services such as AWS to enhance the scalability and efficiency of our system, incorporating the latest research on parameter-updating optimization methods for LLM-based multi-agent systems. This comprehensive approach ensured that our conversational agents not only met the dynamic needs of e-commerce but also stayed at the forefront of technological innovation.