MrInigo Jauregi Unanue

Dr. Inigo Jauregi Unanue is a Postdoctoral Research Fellow at the Telecom Research Institute (TRI) at University of Technology Sydney, specializing in machine learning and natural language processing (NLP). Beyond academia, he brings 10 years of industry experience as a data scientist across diverse domains including finance, property valuation, IoT sensors, and fraud analysis. He is also co-founder and director of Sendatu Therapeutics, an Australian start-up developing AI-driven gene therapies.

Inigo Jauregi Unanue's research is in natural language processing, with a particular focus on large language models (LLMs). His current work is on aligning and steering LLMs: using preference optimisation and representation editing to control what models generate, and building methods that make their behaviour more reliable, controllable and suited to specific needs. This builds on a long-standing interest in how training objectives shape language models, including reinforcement learning rewards, differentiable training pipelines and adversarial training, applied to text generation, summarisation and classification. A central thread is multilinguality. He has worked extensively on neural machine translation, including discourse-aware and attribute-controlled translation and translation for low-resource languages and specialised fields like Basque and biomedical text. He also works on cross-lingual transfer, so that the benefits of language models reach languages beyond English. More recently, he has extended this to multilingual and code-switched speech in the medical domain.

Inigo's research has been published at the leading venues in natural language processing, including ACL, NAACL, EACL, COLING, LREC and the journal Transactions of the Association for Computational Linguistics (TACL). His work has helped improve machine translation for languages and domains with few resources, including Basque and biomedical text. He has also contributed to the community as a co-author of the findings of the WMT Biomedical Translation Shared Task, an international benchmark for evaluating medical translation systems. His methods for training and controlling language models, from fine-tuning translation models with semantic metrics to steering LLMs with preference optimisation and representation editing, address the broader challenge of making language technologies more accurate, controllable and useful across many languages and specialised fields such as healthcare.