• Worked on integrating retrieval-augmented generation (RAG) techniques to improve chatbot capabilities and enhance user interactions.
• Developed scalable systems for deploying machine learning models in production environments.
• Collaborated across teams to explore AI-driven solutions for various fintech applications.

• Contributing to research on advanced interactive AI systems with a focus on enhancing multi-modal communication technologies.
• Exploring innovative AI applications that aim to bridge the gap between human and AI conversational experiences.
• Investigating privacy and security concerns in the development of AI-driven interaction models.

• Conducted research on FinTech for Older Adults, focusing on enhancing financial technology accessibility and usability for elderly users.
• Investigated mHealth with Generative AI, integrating AI to enhance mobile health apps aimed at promoting physical activity and healthier lifestyles.
• Led the development of PAT PAT – an Emotional Support Virtual Pet Application, integrating machine learning to improve user interaction and engagement.

Keyword: Open World Manipulation With Planning, Grasping Learning, Equivariant Model

• Equivariant Models for Robotic Grasping: Developed an advanced robotic grasping model incorporating SE(2) and safe-z prediction, reducing training epochs by 50%.

  • Implemented Equivariant DQN in CartPole, cutting training time in half, using state and image inputs.
  • Pioneered SO(2)-equivariant reinforcement learning techniques, improving grasping tasks and reducing failed attempts by 50%.

• Vision-Language Models for Strategic Grasping: Integrated LLMs for “Open World Manipulation with Planning,” enhancing robot adaptability in dynamic environments. Developed ThinkGrasp, a vision-language grasping system utilizing GPT-4o, achieving a 98.0% success rate in cluttered scenes, outperforming previous methods by over 50%.

  • Built a robust error-handling framework with LangSAM, boosting reliability in real-world experiments.

• Generative Models for Manipulation Policies: Contributed to the development of IMAGINATION POLICY, a generative point cloud model for key-frame manipulation, improving success rates on RLbench tasks by 20%.

  • Proposed a novel multi-task key-frame policy network, increasing generalization to unseen configurations.

In-service period

• Independently developed a VC++-based Database Integration (DBI) tool for data export and import from Access to Oracle, enhancing deployment efficiency by 30%.
• Engineered a Springboot+Vue electronic invoice management system, seamlessly integrated with the hospital system, boosting business efficiency by 80%.
• Assembled and led a development team to create an internal management system tailored to the company’s operations; oversaw key coding, quality control, and project progression.

• Investigated strategic impact factors in soccer teams based on game records analysis.
• Employed social networks and degree centrality for analytical modeling of contest records.
• Revealed that teams prioritize coordination and communication over individual performance through analysis of Huskies and related soccer games.

• Extracted shopping site reviews from Amazon using BeautifulSoup.
• Performed sentiment analysis on the data corpus with Textblob.
• Developed commenter user profiles through Wordcloud visualization.

• Executed public opinion analysis based on news content from Wenzhou News Network.
• Implemented dictionary-based sentiment analysis and structured single news article into a knowledge graph.
• Utilized TextRank4ZH for extraction of news topic words and visualized topic “hotness.”

• Led the studio to recognition as a national innovation and entrepreneurship training program for college students.
• Initiated an online volunteer management platform for the existing party member volunteer system, optimized for mobile devices.
• Managed maintenance and optimization of studio projects, including server-side responsibilities, and contributed to project analysis and testing.