Dat Hong Email & Phone Number

Santa Clara, California, United States

United States

In this role, I developed an algorithm that enhances a large language model (LLM) through knowledge distillation from a more advanced LLM. My work involves leveraging technologies such as Python (Scikit-learn, Pandas, Numpy), LangChain, Prompt Engineering, OpenAI APIs, HuggingFace Transformers, and GPU training. This innovative approach significantly boosted model performance and interpretability, achieving over >= 9.7% improvement compared to baseline models (from 4.1/5 to 4.5/5). The detailed methodology and results will be submitted to the Marketing Science journal within the next month, and a conference version of this work is currently in progress.In collaboration with the Yale School of Medicine, I applied Explainable AI methods to uncover predictive patterns in medical data, which inform patient outcomes. Our model achieved an AUC of 0.895 and an accuracy of 0.902 by combining techniques like XGBoost and SMOTE/ADASYN. We utilized SHAP values to highlight the impact of cytokines in our model. This research, where I am a co-first author, has been published in AACR 2024 and highlights the impact of specific cytokines on myocarditis and immunosuppression.Additionally, I actively contribute to the Marketing group's research initiatives by participating in seminars and mentoring students. I also engage in new research projects focused on natural language processing for marketing problems, enhancing the Marketing group's research community through participation in seminars and events.

United States

I invented and implemented a new search recommender system for PeacockTV, significantly enhancing user experience and engagement. This system achieved a 38% improvement in precision and an 182% increase in quantitative recall, as demonstrated through experimental results. The technologies utilized in this project included Python (Scikit-learn, Pandas, Numpy), Google Cloud, BigQuery, Google Vertex AI, and Streamlit.

United States

I analyzed and built interactive visualizations and predictive models for annual Diversity, Equity, and Inclusion (DEI) reports from various companies, assisting prospective marketing companies in identifying key social patterns and trends. The technologies used in these projects included Python (Scikit-learn, Pandas, Numpy), Google Colab, and HTML/CSS/JavaScript.

Iowa City, Iowa

In addition to doing research, my responsibilities included teaching classes: grading, holding office hours, leading discussion sessions, designing problem solutions, and guiding students. For graduate courses, I taught Deep Learning, Health Data Analytics, and Information Visualization. At the undergraduate level, I instructed courses in Fundamentals of Computer Science, Algorithms, and Discrete Structures.

Iowa City, Iowa Area

I analyzed the HCUP dataset, which contains 27 million records of patient emergency room and hospital discharges. Our goal is to mine knowledge from the dataset and build an accurate classifier to predict if an emergency visit will result in rehospitalization within 7,14 and 30 days. Achievement: We can achieve 82% AUC and are still in the process of improving it.In addition, I developed a cough detection system to classify a sound is a cough sound or not. Our goal is to count how many cough sounds from hundreds of hours of sound collected from the hospital. Achievement: We developed models with Logistic Regression/CNN/LSTM to classify cough with > 96% AUC.

Vietnam

Full-stack Java developer

Full-stack software engineer

Java Web Developer

Doctor Of Philosophy - Phd, Computer Science

Master’S Degree, Computer Science, Major Gpa 3.91/4.0, Accumulated Gpa 3.83/4.0

Bachelor'S Degree, Computer Science And Engineering