- Fine-tuned pre-trained LLMs (GPT-3.5-turbo-0613, HuggingFace BERT-base) on product review datasets to perform sentiment analysis and keyword extraction; integrated transfer learning techniques and optimized hyperparameters using AdamW; improved sentiment analysis accuracy by 82% using cross-entropy loss and keyword extraction F1-score by 38% - Designed a 3-layer LSTM architecture incorporating ReLU activations, dropout regularization, and batch normalization to process user behavioral data (e.g., views, clicks, purchases) and survey features; implemented embedding layers to encode categorical inputs and utilized a time-distributed dense layer for temporal feature learning. - Built a product ranking algorithm using an ensembled multi-classifier approach (Random Forest, Gradient Boosting, XGBoost) on 48 product parameters; optimized hyperparameters with Hyperopt, resulting in a 9% increase in purchase rate and a 38% rise in click-through rate. Deployed the pipeline using TensorFlow, PyTorch and BentoML, incorporating ONNX for cross-platform compatibility and real-time inference scalability.

더보기

- Fine-tuned a GPT-3 Davinci model using PyTorch to generate high-fidelity images from natural language prompts, leveraging advanced transfer learning techniques. - Streamlined machine learning deployment workflows via Azure Machine Learning Studio, reducing deployment latency by 35%. - Engineered a fully operational web app using Node.js for model deployment, integrating OAuth-based social network authentication through Azure App Services to enhance user security and scalability.

더보기

- Built the website user-interface using HTML, CSS, JavaScript and plug-ins/add-ons to build, enhance and troubleshoot the website and its capabilities - Implemented front-end architecture through React.JS to design user-facing features and building reusable components and front-end libraries for easy development in the future - Gained experience with the Node.JS, Git Version Control System and RESTful APIs - Published Website: https://teracomix.com/new

더보기

- Published and authored the peer-reviewed article “Analyzing Consumer Behaviors and Attitudes Towards Plastic Bag Consumption in Hanoi Wet Markets" alongside collaborators from UVA, Duke, UC Berkeley, and TU Delft. - Implemented unsupervised machine learning algorithms, including K-Nearest Neighbors (KNN) and Support Vector Machines (SVM), to analyze stated preference surveys (n=1000+), identifying material composition and cost as the most influential factors in plastic bag usage. - Designed and executed an A/B test involving 1200+ participants across two homogeneous wet markets, finding that a 1,000 VND increase in plastic bag price reduced consumption by 92%, demonstrating the potential efficacy of price-based intervention policies.

더보기

- Conducted research on Laos’s primary education market compared to other Southeast Asian emerging markets, especially in regards to enrolment in private institutions, net enrolment rate and variations across sexes using data from the World Bank and the Laos Government; stakeholders: Montrose International, World Bank, and International Labor Organization - Employed Python’s ‘wbdata’ module and ‘pandas’ library & R’s ‘WDI’ library and ‘ggplot2’ in order to scrape data from the World Bank V2 Indicators API & create visualizations and graphs - Used Excel to perform qualitative statistical analysis for Know One Teach One Vietnam (KOTO) graduates to examine their level of success depending on multiple variables, and its correlation to gender, race and ethnicity; published findings

더보기