My PhD research explored natural language processing, representation learning, and topic models. I continue to value open source as a way to make research useful, reproducible, and easier to build on. I like projects that make difficult ideas easier to test, compare, and use.
OCTIS
OCTIS is an open-source framework for training, evaluating, and comparing topic models. I designed and developed the project to support fairer comparisons through hyperparameter optimization. It has earned more than 800 GitHub stars.
Contextualized Topic Models
Contextualized Topic Models combines contextual language-model representations with neural topic models. I collaborated with the MilaNLP Lab on the project and am a maintainer and contributor. The project has more than 1,300 GitHub stars.
FashionCLIP
FashionCLIP explores vision-language learning for general fashion concepts. I contributed to research and evaluation work associated with the project; the project itself is led and maintained by its wider contributor community. Its models have reached more than 100 million downloads on Hugging Face.
Research interests
I’m interested in representation learning, evaluation, recommendation, and the bridge between research results and production systems. Find my academic publications on Google Scholar, or browse selected talks.