I’m a UCL MEng Computer Science graduate, looking for PhD/RA opportunities.
Current Interests
- Human-like AI voice
- Human-AI natural interaction in conversation
- LLM preferences under conflicts and related refinement
- generic AI&ML stuffs
Research Projects
- Final Year Project: Motion Adaptation via unsupervised DNNsand report
- Meow-Omni 1: A Multimodal Large Language Model for Feline Ethology (under review at NeurIPS 2026)
- When Specifications Conflict: A Symmetry-Based Framework for Measuring LLM Preferences (under review at AAAI 2027)
Currently Exploring
I’m currently interested in how to make AI interacts with humans more naturally, in speech and conversation in particular. One interesting broad question is how AI dialogue system works correctly in real world environment, e.g. how it recognises target user in a noisy environment (not gonna be solved by a simple VAD). There are many directions to be explored, which may directly or indirectly improve the interaction such as representation of semantics, acoustic and paralinguistic, system design (recent Moshi) or even findings in linguistics.
What I am thinking recently can be formulated as a specific research question: Can user-stream routing dynamically adapt to the reliability of incoming speech? This can be considered as a follow-up and combination work of IRAF and Lu et al.