Yiping Li Email & Phone Number

Irvine, California, United States

• Designed and implemented backend infrastructure for OpenAI API integration, streamlining functionalities like assistant interaction, chat generation, function calls, and file search, resulting in a 20% reduction in token usage• Enhanced business features such as job description generation and resume parsing through refined prompt engineering and task decomposition, boosting content generation accuracy from 42% to 95%• Conducted an in-depth analysis of GraphRAG, authoring a comprehensive document detailing its benefits, potential use cases, and implementation workflow; developed and deployed user-friendly interfaces to facilitate business integration

Atlanta, Georgia, United States

• Conducted an in-depth analysis and performed hyperparameter search on the YOLOv8 model tailored to our data, researched runtime and performance trade-offs, and identified the most cost-effective hyperparameters• Proposed and implemented a new document classification method using word frequency analysis, improving accuracy from 0.99 (achieved by YOLOv8) to 1.0 while significantly reducing costs• Replaced the original Oriented Bounding Box (OBB) model with the RANSAC algorithm, significantly improving text extraction accuracy from 32.8% to 99.4% by effectively addressing image rotation and scaling challenges• Architected and implemented the backend of the new version document digitization and data extraction system under the Django framework

Beijing, China

• Conducted an in-depth analysis and performed hyperparameter search on the YOLOv8 model tailored to our data, researched runtime and performance trade-offs, and identified the most cost-effective hyperparameters• Proposed and implemented a new document classification method using word frequency analysis, improving accuracy from 0.99 (achieved by YOLOv8) to 1.0 while significantly reducing costs• Replaced the original Oriented Bounding Box (OBB) model with the RANSAC algorithm, significantly improving text extraction accuracy from 32.8% to 99.4% by effectively addressing image rotation and scaling challenges• Architected and implemented the backend of the new version document digitization and data extraction system under the Django framework

Beijing, China

• Developed the Dynamic Gradient Tailor (DGT) plugin to solve detrimental gradient interference in Federated Learning on non-independently and identically distributed (non-I.I.D) data• Developed a novel federated learning (FL) framework that dynamically adjusts client participation to enhance FL efficiency on unstable distributed data• Conducted research on the dynamic curriculum-based knowledge distillation framework (DCD), to mitigate the damages due to heterogeneous sampling during distillation process• Independently designed and implemented a scalable and maintainable testing framework for Federated Learning, enabling our lab to efficiently run experiments across diverse settings• Conducted extensive experiments and research on aggregating models fed with non-I.I.D data under strong resource constraints, gaining a deeper and clearer understanding of the underlying challenges

Beijing, China

• Proactively collected online videos and created multiple datasets for Deepfake detection evaluation, developing custom scripts for tasks like video-to-image conversion, image-to-video generation, and video clipping/rotation to streamline the process• Developed an evaluation pipeline for Deepfake detection models, featuring dataset sample balancing and automated report generation with metrics like distribution, ROC, and F1 score, which contributed to a 2% increase in accuracy• Collected and curated an extensive adversarial text dataset, encompassing all letter and number substitutions in UTF-8 characters to enhance model robustness against text-based attacks• Developed a rule-based model utilizing the adversarial text dataset, effectively blocking over 600,000 harmful comments that bypassed the BERT model, significantly enhancing content moderation

Master'S Degree, Computer Science

Bachelor'S Degree, App & Comp Math Sci (Sci Comp & Num Al), Psychology