Machine Learning Data Science

Hanwen (Hudson) Xing

Email: ********@***.*** Mobile +1-347*******

Education Background

University of Southern California, Viterbi School of Engineering GPA: 3.91/4.00 California, USA Master of Science in Computer Science Aug 2023 - Dec 2025

• Coursework: Advanced Natural Language Processing, Analysis of Algorithm, Computer Networks, Machine Learning Nankai University, School of Statistics and Data Science Major GPA: 3.82 (90.32) Tianjin, China Bachelor of Science in Statistics and Data Science Sept 2019 - Jun 2023 Columbia University, Exchange Program with scholarship GPA: 4.00/4.00 New York, USA

• Main Coursework: Applied Data Analysis (A+), Stochastic Process (A), Applied Data Mining (A) Jan 2022 - May 2022 University of California, Berkeley, Exchange Program with scholarship GPA: 4.00/4.00 California, USA

• Main Coursework: Artificial Intelligence (A), Reinforcement Learning (A) Aug 2022 - Dec 2022 Internship Experience

Amazon Web Services Santa Clara, USA

AWS SageMaker Department AI/ML Software Development Engineer. May 2025 – Current

• Focusing on optimizing AI infrastructure for scalable generative AI training and deployment on AWS SageMaker AI Platform. TikTok Seattle, USA

Seed Conversational-AI Department Applied Scientist (Off-Cycle Intern) Jan 2025 – May 2025

• Focusing on building Conversational AI - Designed and built a RAG-based intelligent customer service agent pipeline from scratch to enhance e-commerce FAQ response efficiency. Implemented a retrieval-augmented generation (RAG) architecture to improve answer relevance, leveraging clustering algorithms to identify and mining high-value FAQs.

• Incorporated SOTA techniques such as GraphRAG and RippoRAG to construct a semantic document graph using FAQ clustering and entity co-reference. This led to a 7% gain in recall@5 and improved multi-intent query resolution compared to vanilla RAG. Texas Medical Center Houston, USA

Department of Data Science and Artificial Intelligence Machine Learning Engineer May 2024 – Aug 2024

• Optimized the pre-trained model through the finetuning method: LoRA, Prefix-tuning, and prompt engineering with Pytorch, enhancing performance by Llama 3. Conducted comprehensive gene and GO term analysis, developing algorithms to correlate gene expressions with biological terms. Improved the model's predictive accuracy by 26%, achieving a 0.41 ROUGE-L score. (under review)

• Developed a heuristic search method utilizing a finetuned Large Language Model to detect the presence of GO-terms in text segments. Finetuned the LLM on domain-specific datasets to improve its understanding of context with PyTorch. Enhanced the model’s ability to accurately identify relevant GO terms from the target paragraph, thereby improved precision in annotation and related research. KuaiShou Technology Beijing, China

E-commerce Department Machine Learning Engineer Feb 2023 - Apr 2023

• Optimized KuaiShou's recommendation algorithms with NLP techniques like Item2Vec and Word2Vec. Analyzed intricate datasets from 200 million users and over 10 billion videos. Developed and trained a Behavior Sequence Transformer model using TensorFlow. Embraced advanced hyperparameter tuning strategies, optimizing the model's F1 score from 0.76 to an impressive 0.81 and proudly achieving an AUC- ROC score of 0.82. Resulted in a 15% surge in CTR and an 8% boost in purchase conversion rates after A/B testing.

• Adopted Tree-based models like XGBoost and Random Forest to target "persuadable" customers. Processed, cleaned, and enriched datasets from 5 million diverse customer records, achieving an 83% prediction accuracy. Augmented the model's precision by 20%. University of Southern California Los Angeles, USA Viterbi School of Engineering Graduate Teaching Assistant Aug 2024 – Dec 2024

• Instructed two lab sections of 60 students for CSCI455 Programming Systems Design, covering Java/C++ programming, object-oriented design, recursion, inheritance and algorithm analysis. Emphasized good coding practices, unit testing, and debugging. Introduced the Linux environment, shell scripting, and makefiles. Held office hours and assisted in course material development. CSC Financial Co. Beijing, China

Quantitative Investment Department Quant Developer May 2021 - Sept 2021

• Calculated the estimated quantitative strength of the constituents' consistency, built a strategy, and achieved an annual excess return of 20% in backtests and 10% excess return in simulated trading.

• Selected more than 100 monthly macroeconomics variables, used Tensorflow framework to build LSTM-based time series prediction model to predict the next month’s return of the HS300 index, used Adam optimizer to train the model and time-series-based ordering cross validation to select hyperparameters, backtesting return is above 800% within 10 years. Project and Research Experience

Offline Reinforcement Learning Optimization Project at UC Berkeley Berkeley, USA Advisor: Prof. Sergey Levine Aug 2022 – Apr 2023

• Studied how offline reinforcement learning models adapt as offline dataset sizes grow during training. Using pre-trained policies, gathered initial datasets from four MuJoCo environments. Expanded this dataset with additional rollouts from current training policies in these environments. Using overestimation error for evaluation, enhanced optimization of the data collection method with added rollouts.

• Innovatively introduced a small amount of online data to offline reinforcement learning to bridge the distribution gap between data collection strategies and algorithmically trained policies. Implemented Implicit Q-Learning (IQL) on top of Conservative Q-Learning (CQL) to analyze expected loss functions. Significantly, leveraging IQL led to a 15-fold reduction in overestimation errors induced by distributional shifts. Yale’s Summer Research in Zhao’s Lab New Haven, USA Advisor: Prof. Hongyu Zhao Apr 2022 – Sept 2022

• Evaluated traditional PRS methods under various simulations and assumptions, then designed deep neural networks, including CNN and RNN, for PRS development and compared them with Bayesian approaches.

• Introduced a new nonparametric dynamic network model tailored for multiple dynamic networks with key features like low-rank structure and structural sparsity. Used group lasso for sparsity, devised a block gradient descent algorithm to enhance tensor factorization. Handwritten Chinese Characters Recognition Research at Columbia University New York, USA Advisor: Prof. Wayne Lee Feb 2022 – May 2022

• Applied grayscaling, size rescale, and Gaussian blur to process images with 5000 handwritten similar Chinese characters, selected features through Hough transform, and Harris corner detection. Implemented 3 machine learning models to predict the correct versus wrong characters: logistic regression, random forest, and finally chose the neural network model with the highest f1-score of 0.93.

• Introduced a new dataset containing different-looking characters and verified the robustness of our final model. Skills and Interests

• Language: Python, R, Java, C++; Cloud: AWS, Azure, GCP; Machine Learning: Sklearn, PyTorch, TensorFlow; Visualization: Tableau, PowerBI, EXCEL; Big Data: Hadoop, Spark; Databases: SQL, MySQL, MongoDB

• Fun FactsJ: Relished sports and electronic gadgets. Proficiency in SLR and drones. Obtained the OW certificate of a scuba diver at 10

Contact this candidate