I am a final-year PhD researcher in Natural Language Processing at the University of Edinburgh and the University of Cambridge through the European Lab for Learning and Intelligent Systems (ELLIS) Network, advised by Shay Cohen, Anna Korhonen, and Edoardo Ponti. I am a recipient of the Apple Scholars in AIML PhD Fellowship, recognizing emerging leaders in AI/ML research, and a member of the ELLIS PhD Cohort, a pan-European network for excellence in artificial intelligence.

My research investigates whether large foundation models (LFM; e.g., LLM and VLM) develop a latent world models across modalities such as text, vision, and speech. Latent world modelling refers to a model’s intrinsic understanding of the environment and agent dynamics, thus enabling reasoning and planning without an external, specialized world model. Around this, my interest spans:

1. World modelling as an emergent capability — eliciting LFM to simulate future states and causal dynamics, with the novel training paradigm requiring minimum supervision. (🌀 SWIRL, BoostrapWM)

2. Language grounding — examining how abstract linguistic representations connect to physical and perceptual reality (e.g., spatial/temporal system) in LFM. (⏳ TemporalGrounding, SpatialGrounding)

3. Next-generation model architecture — advancing unified architecture for the omnimodal FM with any-to-any modality, serving as the substrate for emergent latent world models (💭 vLCM).

4. Hallucination mitigation — designing interventions and evaluation frameworks to diagnose and reduce hallucinations in LFM. (🍄 SEA, mFACT, ICR2, TWEAK)

I have multiple first-author publications at leading AI venues, including ICLR, NeurIPS, ACL, EMNLP, and NAACL. My research is complemented by industry collaborations with Meta FAIR, Apple AIML, NVIDIA Research, and Baidu NLP, spanning both fundamental and applied foundation model research.

I am on the 2026 job market for industrial research scientist positions. I am actively looking for positions based in 🇨🇳/🇬🇧/🇨🇭/🇺🇸/🇪🇺/🇸🇬.