I currently work as a Machine Learning Engineer at TikTok-Data. My job focuses on video content understanding and mutli-modal LLM training.
Previouly, I obtained master’s degree from Language Technologies Institute (LTI) Carnegie Mellon University. I conduct research with Prof. Lei Li and Prof. Maarten Sap on Retrieval-Augmented Generation and Reinforcement Learning. I earned my B.Eng. in Computer Science and Technology from the Chu Kochen Honors College, Zhejiang University. I did research on time series classification supervised by Prof. Yang Yang. I also had wonderful research experiences in CHAI lab with Prof. Chenhao Tan at UChicago about evaluating long-form multimodal summarization generated by LLMs.
🔥 News
- 2026.01: I start a new job as Machine Learning Engineer at TikTok.
- 2026.01: 🎉🎉 One paper about RAG robustness on linguistic variations was accepted by EACL 2026 (Main Conference)!
- 2025.12: I obtained Master’s Degree from LTI, CMU.
- 2025.05: I start Machine Learning Scientist Internship at TikTok.
- 2025.03: I start a new research on time-sensitive RAG benchmarking at LTI under the guidance of Prof. Lei Li!
- 2024.09: 🎉🎉 One paper about segmented time series classification was accepted by NeurIPS 2024!
- 2024.09: I start a new research on RAG’s robustness on linguistic variations at LTI under the guidance of Prof. Maarten Sap!
- 2024.08: I start my master’s program at School of Computer Science, Carnegie Mellon University!
- 2024.07: 🎉🎉 One paper about multimodal long-form summarization was accepted by COLM 2024. My first-author paper!
📝 Publications
Out of Style: RAG’s Fragility to Linguistic Variation
Tianyu Cao*, Neel Bhandari*, Akhila Yerukola, Akari Asai, Maarten Sap
Our research reveals that linguistic variations significantly impact both retrieval and generation stages. RAG systems exhibit greater sensitivity to such variations compared to LLM-only generations, highlighting their vulnerability to error propagation due to linguistic shifts.
RARE: Retrieval-Aware Robustness Evaluation for Retrieval-Augmented Generation Systems
Yixiao Zeng, Tianyu Cao, Danqing Wang, Xinran Zhao, Zimeng Qiu, Morteza Ziyadi, Tongshuang Wu, Lei Li
RARE is a unified framework designed to automatically generate synthetic, dynamic, and time-sensitive corpora for testing Retrieval-Augmented Generation (RAG) systems using domain-specific unstructured datasets.
Characterizing Multimodal Long-form Summarization: A Case Study on Financial Reports
Tianyu Cao, Natraj Raman, Danial Dervovic, Chenhao Tan
Developed an evaluation framework for LLM-generated multimodal long-form financial report summaries, integrating textual and numeric analysis.
Con4m: Context-aware Consistency Learning Framework for Segmented Time Series Classification
Junru Chen, Tianyu Cao, Jing Xu, Jiahe Li, Zhilong Chen, Tao Xiao, Yang Yang
Con4m is a consistency learning framework, which effectively utilizes contextual information more conducive to discriminating consecutive segments in segmented TSC tasks, while harmonizing inconsistent boundary labels for training.
📖 Educations
- 2024.08 - 2025.12, Master, Language Technologies Institute (LTI), Carnegie Mellon University
- 2020.09 - 2024.06, Undergraduate, Computer Science and Technology, Chu Kochen Honors College, Zhejiang Univeristy
💻 Internships
- 2025.05 - 2025.08, TikTok Inc., San Jose, CA