Tianrui Song

Email: T.Song-12@sms.ed.ac.uk

Website: https://tianrui-song717.github.io/

Research keywords: Large Language Models, Multimodal Large Language Models, Agentic AI, Reasoning, AI for Creative Workflow

Bio:

Tianrui Song, who also goes by Ray, is a PhD student in the Designing Responsible Natural Language Processing CDT at the University of Edinburgh. His research interests lie at the intersection of multimodal large language models, agentic AI, responsible AI, and human–AI collaboration. He is particularly interested in building intelligent systems that can understand and reason over long, various and diverse multimodal contexts that support people in complex real-world video production workflows.

Before joining Edinburgh, Tianrui completed an MRes in Artificial Intelligence and Machine Learning at Imperial College London and an MPhil in Artificial Intelligence at HKUST (Guangzhou). His previous research has covered LLM safety alignment, LLM-enhanced recommender systems, and LLM agents.

Alongside his research, Tianrui is a long-time video creator. This experience motivates his interest in creator-centred AI: developing reliable, transparent, and controllable AI tools that can assist with organising, retrieving, reasoning over, and editing video content while preserving human creativity and control.

PhD research:

Tianrui’s PhD research explores creator-centred multimodal AI systems for real-world video workflows. I am interested in how multimodal large language models and AI agents can understand long-form and personal video footages, reason over project context, and collaborate with creators throughout all stages in the video production process.

The research will investigate narrative-aware video representations, creator-adaptive retrieval and recommendation, and agentic editing assistants that can respond to natural-language requests, use tools, and make context-aware suggestions. A central goal is to study how these systems can be made responsible and useful in practice, including how they preserve human control, authorship, transparency, and trust.

More broadly, I am interested in connecting advances in multimodal reasoning and agentic AI with human-centred design to build AI systems that support, rather than replace, creative decision-making.

Supervisors: Frank Keller