Resume
Experience
2023.3 - 2023.12
Capstone Researcher
Inference Analytics, Inc.
Chicago, IL, US
-
Instruction tuned LLMs (LlaMa, Mistral, etc.) with qLoRA to automate radiology report generation for radiologists.
-
Aligned LLMs with RLHF and achieved 0.32 ROGUE-L score. Generated reports were 30.0% factually correct, evaluated by opensource LLMs.
-
Operationalized LLMs with GCP Vertex AI and wandb. Deployed with Hugging Face and Gradio.
2023.6 - 2023.8
Data Science Intern
SAS
Shanghai, China
NLP-based Banking Customer Complaint Auto-classification System
-
Applied unsupervised LDA topic modeling to create 20+ secondary-level labels for the subdivision of 4 core business areas.
-
Employed various deep learning techniques (Bi-GRU, Attention) in TensorFlow to achieve 93.2% multi-class classification accuracy, an 8.8% increase from baseline, and saved 60.0% time over manual classification.
-
Incorporated KS-Test on word embeddings for ongoing monitoring and adaptation to changing customer concerns.
-
Collaborated with the operations team to enhance business processes, resulting in a 12.5% reduction in customer complaints.
Reinforcement Learning-based Joint Debt Collection Strategy Optimization
-
Constructed a data pre-processing pipeline and converted 30M+ rows of tabular data into a Markov Decision Process dataset.
-
Trained reinforcement learning algorithms (CQL, DQN) to develop optimal collection strategies for each joint debt condition.
-
Implemented an A/B test, leading to a 6.7% increase in average collections and a 10.2% reduction in debt write-off rate.
2021.6 - 2021.8
Data Analyst Intern
Industrial and Commercial Bank of China
Shanghai, China
-
Addressed dataset imbalance with SMOTE and ADASYN resampling methods to enhance model performance.
-
Produced 5+ reports with pivot tables and Power BI, illustrating timely and geographical patterns and trends in fraud cases.
-
- Trained machine learning models (XGBoost, Light GBM, etc.) to classify and early detect fraudulent accounts, improving recall by 8.3% and precision by 7.5%.
2021.1 - 2021.3
Corporate Venture Capital Analyst Intern
HP
Constructing a predictive model to gauge the profitability potential of start-up investments.
-
Curated and filtered 3000+ VR start-up data from Crunchbase, exporting records to CSV format for analysis.
-
Identified 6 significant predictors from 20+ variables using correlation analysis and Random Forest (SHAP values).
-
Developed a profitability prediction model using robust regression, achieving an R-squared value of 0.64. Identified five start-ups through the model that received $120K funding on average afterward.
Education
2022 - 2024
University of Chicago
M.S. in Applied Data Science
Chicago, IL
Related Courses
-
Machine Learning and Predictive Analysis
-
Data Engineering Platforms
-
Big Data Platforms
-
Data Mining
-
Time Series Analysis
2018 - 2021
Emory University
B.A. in Mathematics and Economics
Atlanta, GA
Related Courses
-
Advanced Financial Markets
-
Econometrics
-
Data Visualization
-
Mathematical Economics
-
Linear Algebra
Professional skillset
Programming language: SQL, Python (Pandas, NumPy, TensorFlow, D3rlpy), R, Tableau, NoSQL, Spark, Docker, GCP, SAS, Linux
Technical skills: Data Mining, Cloud databases, Machine Learning, Natural Language Processing, Deep Learning, Reinforcement Learning, Times series Analysis, What-if Analysis, Unsupervised Learning, Root-Cause-Analysis, A/B Experiment, Metrics Definition
Languages
English (proficient)
Chinese (native)