• Developed a pipeline to automate the scraping, organization, and preprocessing of about 200,000 academic papers.

• Applied Llama 3.1 70B to classify gene-disease pairs from papers, achieving 80% accuracy and 95% near accuracy.

• Produced visualization software to compare performances of models, allowing rapid prompt adjustments and testing.

VIDA Center

• Implementing a pipeline to generate descriptions for multimodal datasets automatically with LLMs and VLMs.

• Working under the supervision of Prof. Juliana Freire and Ph.D. student Yu-Rong Liu.

• Tailored a product inventory detection program to provide real-time shelf restocking alerts by designing a U-Net model architecture using TensorFlow, collecting and annotating marketplace photos for training data, model training, implementing detection logic, and deploying the program on in-store Raspberry Pi 3.

• Constructed a footfall detection program with PyTorch to track the number of people passing by each shelf by selecting the You Only Look Once version 4 (YOLOv4) model architecture, creating footfall calculation algorithms, and deploying the program on in-store Raspberry Pi 3.

• Revamping an outdated merchant-supplier fruit and vegetable ordering and delivery system.

Lab of Dr. Huei-Ling Chen

• Evaluated the relationship between stock price forecasts from hundreds of analysts and the actual performance of the Chinese stock market using financial metrics, employing Pandas and Numpy.

• Extracted stock price history and analysts’ predictions by crafting web crawlers with Selenium and Requests.