Yipeng Liu

Data Scientist @Sunia

CA

Developed and maintained large-scale web crawlers using Scrapy, Selenium, and BeautifulSoup to collect high-quality text, formula, and table data for OCR model training.· Designed and implemented automated layout data generation, applying 15 transformations (scaling, rotation, shearing, translation, etc) and content filling using OpenCV, PIL, and NumPy.· Built and optimized table data generation pipelines, including layout creation combined with Python and JavaScript. Fine-tuning LLMs (GPT, Qwen) for accurate content translation.· Developed synthetic WhatsApp-style datasets with randomized formatting and content to enhance model robustness.· Evaluated OCR model performance, analyzed data distribution and balance to identify weaknesses, and refined data generation, annotation, and processing pipelines to improve model accuracy and robustness.· Led data annotation efforts, ensuring high-quality labeled datasets for machine learning model development.