CV

Updated in Aug 2024
English Resume
Chinese Resume

Education

Master in Advanced Computing. Tsinghua University, Beijing, CN (Sep 2023 – Now).
Bachelor of Engineering in Automation. Shanghai Jiao Tong University, Shanghai, CN. (Sep 2017 – June 2021).
GCE A-Levels. Tunku Abdul Rahman University of Management and Technology, Kuala Lumpur, Malaysia (2015.05 – 2016.11).

LLM Algorithm Engineer Intern (Jul 2024 - Now)
Zhipu AI, Beijing, China
- Agentic RAG
  - Conducted a comparative analysis of the company’s existing RAG framework against QWEN-Agent and the long-context model GLM-9B-128k.
  - Prepared evaluation scripts and datasets, including HotpotQA and NQ, augmented with additional distractors to increase difficulty.
AI Engineer (Nov 2021 - Aug 2023)
Wise AI, Kuala Lumpur, Malaysia
- Face Anti-Spoofing
  - Led the R&D efforts for the development of a face anti-spoofing model.
  - Achieved compliance with ISO-30107 Level 1 Presentation Attack Detection Standards.
  - Managed the entire project lifecycle, including data collection, model design, training, evaluation, and enhancement.
- Digital Avatar
  - Conducted R&D on the technologies behind digital avatars and broke them down into manageable components based on resource constraints.
  - Conducted in-depth research on topics such as talking head and voice cloning, contributing to the team’s expertise.
  - Led the development of a functional prototype of a digital avatar with voice cloning abilities.

DiaKoP: Dialogue-based Knowledge-oriented Programming for Neural-symbolic Knowledge Base Question Answering (CIKM demo track 2024)
- Developed a dialogue-based knowledge-intensive question answering system.
- Designed and implemented both the frontend and backend of the system.
  - Frontend: developed based on Gradio to facilitate user interaction.
  - Backend: developed based on the FastChat framework to support model deployment, includes modules such as dialogue history tracking, natural language understanding, dialogue policy, and knowledge retrieval strategies.
Text-to-UI: Text-to-User Interface
- Collected datasets from the web comprising pairs of ``User Requirements - Tailwind CSS”.
- Fine-tuned the CodeLlama model using DeepSpeed with the LoRA strategy. The fine-tuned model generates corresponding frontend interface code as output when provided with user requirements in natural language.
MRI Brain Image Retrieval using GNNs
- Modeled the relationships between images using Graph Neural Networks (GNN), formulate the retrieval task into graph node classification tasks.
- Compared to baseline models, the GNN-based method improved retrieval accuracy by 2-8\%.