MPhil student in Data Science and Analytics Thrust, Information Hub, HKUST(GZ)

2023 cohort Red Bird MPhil student

Tencent - Interactive Entertainment Group (IEG) • RAG-based LLM for Private Domain Intelligent Assistant: Led the design and implementation of LLM upgrades for the Game Companion Intelligent Assistant, improving query resolution across multiple platforms (games, WeChat, QQ). Developed a dual-path system to handle highly customized game operations queries through traditional Q-Q matching, with RAG managing more complex game content queries, reducing maintenance burdens. • RAG-based LLM for Game Intelligent NPCs: Contributed to the design and implementation of a multilingual intelligent NPC system for Tencent’s overseas games. Developed an automated evaluation pipeline using multilingual-e5-large to support Chinese, English, and Arabic, significantly improving GSB metrics through refined bad case handling and model adjustments with full weight SFT by Deepspeed framework on Tencent Cloud GPU clusters.

• Managed TCL’s semiconductor display business, centralized document management, and developed an LLM-based intelligent manufacturing application for equipment fault handling, improving efficiency with LLM knowledge assistant.• Prepared and embedded external guides and manuals into vector embedding knowledge bases using multilingual-e5-large models. Tested and optimized LLM for topic recognition and content summarization.• Tested and selected base models, including Grok-1, DBRX, Command-R, Qwen1.5-72B-chat, for deployment.• Developed multimodal data recognition module, converting table data and irregularly formatted text from PDFs into usable text files.• Conducted research on structural data processing and implemented a multimodal table question-answering system.

Be responsible to collect and calibrate the data and sensors of test vehicles and make them workable for AI trainning purposes in XC division of Bosch Automotive Products (Suzhou) Co., Ltd.