People / Students / Class of 2027
Weiyi (Karina) Tian is interested in applying machine learning, statistical modeling, and data science methods to solve practical problems across business analytics, intelligent systems, and biomedical applications. Her current research interests include time-series forecasting, large language models, retrieval-augmented generation, predictive modeling, and interpretable machine learning for decision-making. She completed her undergraduate studies at the University of California, Santa Barbara, where she double-majored in Statistics & Data Science and Mathematics with a concentration in Applied Mathematics.
Karina gained hands-on industry experience as a Data Scientist Summer Intern at BONC in Beijing, where she worked on forecasting, retrieval-augmented generation, and large language model fine-tuning projects. She developed SARIMA and Prophet forecasting models for a major telecom operator, improving revenue forecasting accuracy and generating insights on revenue growth, seasonality, and budgeting. She also built and deployed an on-premise RAG system using LangChain, FAISS, and Qwen2.5-3B, supporting local data security while reducing latency and external token costs. In addition, she implemented an LLM fine-tuning workflow with Qwen2.5-0.5B-Instruct, applying LoRA to improve training efficiency and building automated ROUGE/BLEU evaluation pipelines for model benchmarking.
Karina has also explored machine learning applications in biomedical and health-related research, including predictive modeling projects related to diabetes and heart disease. Through these projects, she worked with clinical and health datasets to identify important risk factors, compare model performance, and examine how data-driven methods can support early detection and risk assessment. At Northwestern, she hopes to further strengthen her foundation in machine learning, deepen her experience with real-world data science projects, and develop more interpretable and scalable models for decision-making across domains. She looks forward to learning from the MLDS community while continuing to build tools that make complex information more actionable and useful.
