Education
HKUST(GZ)
Ph.D. Candidate at the Data Science and Analytics Thrust, supervised by Prof. Nan Tang and co-supervised by Prof. Yuyu Luo.
AI + Data Research · FinTech Entrepreneurship
I build trustworthy AI systems for data and decision intelligence.
Ph.D. Candidate at HKUST(GZ), optimizing Large Language Models for data transformation and financial markets.
Research
Good Data for AI
Data cleaning, transformation, and benchmarking for reliable AI systems, including MegaTran.
Finance
Good AI for Funds
Live evaluation and trustworthy decision support for fund investment intelligence through DeepFund.
Venture
Paradoox AI
Founder and CEO of a FinTech startup focused on trustworthy AI solutions for fund investment.
Research
Publications
arXiv 2025 2025 AI / LLMs
Rise of the Community Champions: From Reviewer Crunch to Community Power
Changlun Li, Yao Shi, Yuyu Luo, Nan Tang
arXiv
Abstract
A study of how community participation and reviewer incentives can reshape overloaded scholarly review workflows and help scientific communities scale peer assessment.
@misc{li2025communitychampions, title={Rise of the Community Champions: From Reviewer Crunch to Community Power}, author={Li, Changlun and Shi, Yao and Luo, Yuyu and Tang, Nan}, year={2025}, eprint={2503.18336}, archivePrefix={arXiv}} IJCAI 2025 2025 AI for Finance
Will LLMs be Professional at Fund Investment? DeepFund: A Live Arena Perspective
Changlun Li, Yao Shi, Yuyu Luo, Nan Tang
International Joint Conference on Artificial Intelligence, FinLLM Workshop
Abstract
This work frames fund investment as a live arena for evaluating whether LLMs can act as professional investment agents, emphasizing real-time decision quality, robustness, and transparent benchmarking.
@inproceedings{li2025deepfundarena, title={Will LLMs be Professional at Fund Investment? DeepFund: A Live Arena Perspective}, author={Li, Changlun and Shi, Yao and Luo, Yuyu and Tang, Nan}, booktitle={IJCAI FinLLM Workshop}, year={2025}} NeurIPS 2025 2025 AI for Finance
Time Travel is Cheating: Going Live with DeepFund for Real-Time Fund Investment Benchmarking
Changlun Li, Yao Shi, Chen Wang, Qiqi Duan, Runke Ruan, Weijie Huang, Haonan Long, Lijun Huang, Nan Tang, Yuyu Luo
Neural Information Processing Systems
Abstract
DeepFund introduces a live, real-time fund investment benchmark that avoids temporal leakage and evaluates LLM-driven investment agents under market conditions closer to deployment reality.
@inproceedings{li2025deepfundlive, title={Time Travel is Cheating: Going Live with DeepFund for Real-Time Fund Investment Benchmarking}, author={Li, Changlun and Shi, Yao and Wang, Chen and others}, booktitle={NeurIPS}, year={2025}} VLDB 2025 2025 AI / LLMs
Weak-to-Strong Prompts with Lightweight-to-Powerful LLMs for High-Accuracy, Low-Cost, and Explainable Data Transformation
Changlun Li, Chenyu Yang, Yuyu Luo, Ju Fan, Nan Tang
International Conference on Very Large Data Bases
Abstract
MegaTran studies weak-to-strong prompting strategies that use lightweight models to guide more powerful LLMs, improving the accuracy, cost profile, and explainability of data transformation pipelines.
@article{li2025megatran, title={Weak-to-Strong Prompts with Lightweight-to-Powerful LLMs for High-Accuracy, Low-Cost, and Explainable Data Transformation}, author={Li, Changlun and Yang, Chenyu and Luo, Yuyu and Fan, Ju and Tang, Nan}, journal={PVLDB}, year={2025}} arXiv 2024 2024 AI / LLMs
SketchFill: Sketch-Guided Code Generation for Imputing Derived Missing Values
Yunfan Zhang, Changlun Li, Yuyu Luo, Nan Tang
arXiv
Abstract
SketchFill explores sketch-guided code generation as a practical path for imputing derived missing values, combining user intent with LLM-generated transformation logic.
@misc{zhang2024sketchfill, title={SketchFill: Sketch-Guided Code Generation for Imputing Derived Missing Values}, author={Zhang, Yunfan and Li, Changlun and Luo, Yuyu and Tang, Nan}, year={2024}, eprint={2412.19113}, archivePrefix={arXiv}}Artifacts & Engineering
Open research systems, live benchmarks, and FinTech codebases.
A bento grid of reusable engineering artifacts behind the research agenda.
A live arena for evaluating LLMs in fund investment.
DeepFund benchmarks investment agents with real-time market constraints, reducing time-travel leakage and making evaluation closer to deployment.
PythonPyTorchTransformersLLM AgentsFinance
Weak-to-strong LLM prompting for data transformation.
An open research codebase for reliable data cleaning and transformation, pairing lightweight and powerful LLMs for cost-aware accuracy.
PythonLLMsData SystemsPrompting
Trustworthy AI products for fund investment research.
A FinTech venture translating research on LLM evaluation and investment intelligence into practical decision-support systems.
FinTechTrustworthy AIAnalyticsLLM Evaluation
Sketch-guided code generation for derived missing values.
A research artifact that uses sketches as controllable intermediate representations for LLM-generated imputation logic.
PythonCode GenerationData ImputationLLMs
Experience
Research and engineering trajectory.
2025 - Present
Founder & CEO
Paradoox AI
Building trustworthy AI products for fund investment.
2024 - Present
Ph.D. Candidate
HKUST(GZ), Data Science and Analytics Thrust
Applied AI, LLM data systems, and trustworthy investment intelligence.
2020 - 2024
Data Scientist
Measurable AI
Derived consumer insights from transactional data across emerging markets.
2016 - 2020
B.Eng. Information Engineering
The Chinese University of Hong Kong
Minor in Computer Science.