About Me
I completed my PhD in Computer Science at the University of Liverpool, supervised by Prof. Danushka Bollegala and Dr. Bei Peng. I previously obtained an MSc in Data Science and Machine Learning with Distinction from University College London and a First-Class BSc in Artificial Intelligence from the University of Liverpool.
My research focuses on large language models, retrieval-augmented generation, synthetic data, trustworthy evaluation, and agentic AI. I study how generative models use evidence, adapt to new data, and can be evaluated through reproducible and human-grounded methods.
Publications
Projects
Evaluation Platform for Agentic LLMs
- Designed and implemented an end-to-end, multi-provider platform for auditing search-enabled and agentic LLMs
- Integrated automated data collection, source and evidence enrichment, evaluation metrics, human review, dashboards, testing, and Linux deployment
- Evaluated source selection, evidence use, provider differences, and repeated-run stability
CommonSyn: Synthetic Data for Diversified Reasoning
- Developed a generation-and-selection pipeline for constructing diverse, high-quality commonsense training data
- Fine-tuned 11 Llama, Qwen, and Gemma models using supervised fine-tuning with LoRA
- Evaluated quality, diversity, transfer to unseen tasks, and retention of general reasoning capabilities
Retrieval-Augmented Generation and Social Bias
- Evaluated how retrieved evidence changes social bias across 16 LLMs, three languages, and four bias categories
- Separated the effects of document collections, retrievers, prompts, and generator models
- Compared mitigation strategies including prompting, evidence summarisation, and Direct Preference Optimisation
Companies Knowledge Graph Construction
- Design the schema of the knowledge graph
- Crawled the company information from websites
- Develop algorithms to fetch the inference chains
Financial Open Information Extraction
- Construct knowledge graph by designing a BERT open information extraction system
- Create the dataset from Wikipedia and SEC filings
Abstract Summarization
- Applying 3 pre-trained models (BERT, MASS and MiniLM) on abstract summarization task and compared the performance on WikiHow dataset
Negative Transfer of Word Meta-Embedding
- Research on negative transfer of word embedding under supervision of Prof. D. Bollegala
- Show great performance on tasks such as semantic similarity
Search Filter
- Developing a website that answers user's questions by crawled webpages
- TextRank algorithm and cosine similarity to rank the paragraphs and webpages
Education
PhD in Computer Science
University of Liverpool, UK
MSc Data Science and Machine Learning
University College London, UK
Distinction Degree, 75% Average Score
BSc Artificial Intelligence
University of Liverpool, UK
First Class Degree
Experience
Service
Reviewer
2024 – PresentACL, EMNLP, EACL, IJCNLP, COLING
Teaching
Teaching Assistant
2022 – 2023University of Liverpool
- COMP337/527 Data Mining and Visualization, 2023
- COMP304 Knowledge Representation, 2022/2023
- COMP518 Database and Information System, 2023
Internships
Knowledge Graph Developer
Jan 2022 – Sep 2022IICT, Suzhou
- Researched NLP algorithms and knowledge graph on hot news and company information
- Constructed and maintained a finance knowledge graph
- Tracked, studied, reproduced, and improved up-to-date machine learning methods
Webpage Developer
Aug 2019 – Jun 2021Xi'an Jiaotong Liverpool University, Market Department
- Implemented and maintained a market customer analysis website
- Helped staff to analyze customer's preferences with their job, province and university
Software Engineer
Jun 2017 – Sep 2017RISE English School
- Developed a system to manage the students and classes
- Helped teachers to easily track their students' courses, classes and tuition fee
- Updated the information automatically in the database and generated weekly reports