KEWEI LIU
Master's Student | Multimodal AI · Speech & Language Processing
Tokyo, Japan • liu-kewei@gavo.t.u-tokyo.ac.jp • github.com/kayui-gavo • website
EDUCATION
Minematsu–Saito Laboratory. Current research on AI-assisted speaking support for Japanese learners.
Yoshii Laboratory. Bachelor's thesis on user-controlled multimodal video summarization.
RESEARCH & PRODUCT DEVELOPMENT
- Defined a video-summarization task in which users specify how the emotional tone should evolve while the summary still retains key scenes.
- Designed a multimodal model that adjusts image, audio, and subtitle contributions scene by scene and selects a sequence that balances the requested progression with content importance.
- On TVSum (50 YouTube videos, 5-fold cross-validation), improved internal trajectory alignment by 24% (0.427 to 0.531) while maintaining comparable overlap with human-rated key scenes. Led problem formulation, implementation, evaluation, analysis, and manuscript preparation.
- Built speech-evaluation logic for Chinese-speaking learners of Japanese, producing separate feedback on pronunciation, rhythm, fluency, and pitch.
- Diagnosed cases where mismatched or low-quality recordings received implausibly high scores and introduced a no-score policy when the available evidence was insufficient.
- Used AI coding assistants to compare implementation options and draft edge-case tests, then validated changes with audio review, structured logs, automated tests, and a browser-based failure-review tool.
PROFESSIONAL EXPERIENCE
- Work on education research, curriculum and learning design, teaching materials, educational content, and education product development for students preparing for Japanese university admissions.
- Design physics curricula and materials for the EJU, Common Test, and university entrance examinations, informed by classroom practice and recurring learner difficulties.
- Taught mathematics and physics to international students and advised on school selection, application planning, personal statements, interviews, and oral examinations.
- Used student questions and written errors to identify recurring barriers in reading Japanese problems, drawing physical models, and setting up equations; revised lessons and exercises accordingly.
ADDITIONAL EXPERIENCE & RECOGNITION
Kyoto University: Electrical & Electronic Circuits OA, Oct. 2024 – Mar. 2025; International Student Tutor, Apr. 2025 – Mar. 2026.
Selected internships (2026): Sony Corporation — Internship; Rakuten Group, Inc. — Software Engineering Internship.
Awards & Scholarships: Ichikawa International Scholarship Foundation Scholar (2023–2024); MEXT/JASSO Honors Scholarship (2022–2023); Excellence Award, Electrical and Electronic Circuits Poster Presentation, Kyoto University (2023).