Lead applied data scientist team on conducting NLP related projects.

• Enhanced Skill Taxonomy: Leveraged sources like Wikidata to expand the skill taxonomy, incorporating additional attributes such as job roles, educational backgrounds, and categorical classifications for each skill item.• Advanced Skill Extraction: Evaluated and implemented various Named Entity Recognition (NER) models, including GLiNER and LLM‑based approaches, surpassing state‑of‑the‑art (SOTA) performance on the Skillspan dataset for skill extraction accuracy.• Optimized Word Similarity Tasks: Investigated large language models (LLMs) for word similarity tasks, achieving performance comparable to specialized models and utilizing Retrieval‑Augmented Generation (RAG) to improve prediction precision.• Developed Comprehensive Demo: Integrated all skill‑related services into a unified demonstration framework, facilitating public reviews and showcasing feature capabilities effectively.

• Transformed Skill Taxonomy: Led the conversion of a manually maintained skill taxonomy into a unified, structured data format by leveraging augmented information and Large Language Models (LLMs).• Developed Merging Algorithm: Designed and implemented an algorithm that combines embedding and cross‑encoding techniques to merge similar items, incorporating human feedback for verification.• Fine‑Tuned In‑House LLMs: Successfully fine‑tuned proprietary LLMs with human‑annotated data, achieving superior performance compared to large commercial models (e.g., Llama3 8b, phi3 mini‑4k,Qwen2 0.5b, and 1.5b) while maintaining reduced model sizes.• Enhanced Model Capacity: Boosted model capabilities by integrating diverse task conditions, contributing to the overall success and efficiency of project objectives.

• associated 40+ interdisciplinary research projects, such as the emotional expression of animated characters and people’s cognition under tactile interruptions (haptics through the controller) in VR (virtual reality)• effectively guided clients through the analysis stage, motivated clients based on the win‑win strategy and educate clients with advanced statistical models, such as generalized linear model, random effects, and ordinal data type• validate experiment designs and survey questionnaires, unveil data patterns through diverse visualizations(ggplot2, seaborn)• Provide demonstrations and technique supports for statistical platforms: R, SAS, SPSS, and JMP

• Set up Linus environment in GPU clusters and evaluate the weakness of deep neural networks under PyTorch framework• Adopted 10+ image processing methods using OpenCV & scikit‑image to expand the learning space of the neural networks• Experimented with Reinforcement Learning in exploring model parameter space to improve the model robustness• Investigate the image flow passing through each network layer and distinguish the contaminated images from their embedded feature space using Unsupervised Clustering (t‑SNE, UMAP)• Developed a toolkit for comparing real‑time performance of CNN under various image filters (”Model Eval” at IOS App Store)

• examined the association between the class label and topological features in the 3D ShapeNet dataset• constructed the time series graphic representation of the Ethereum Blockchain transactions, proposed a novel analysis method to monitor market status and detect abnormalities

Assisted in products Finite Element Analysis Studies. Verify product feasibility and capacity under industrial standards.