Dimo Angelov

WORK HISTORY

EDUCATION

SKILLS

ABOUT DIMO ANGELOV

I build AI systems that discover how markets are actually structured.I’m an AI researcher and practitioner specializing in vector embeddings, unsupervised learning, information retrieval and language models. My work focuses on uncovering hidden structure in complex data – from text corpora to real-world market dynamics – and turning that structure into usable intelligence.I created Top2Vec, a widely adopted open-source algorithm for topic modeling and semantic search (3k+ GitHub stars), used across NLP, social science, and biomedical research. I later published Contextual-Top2Vec (EMNLP 2024), extending this work into contextual and multi-vector representations.I hold a PhD in AI with a focus on NLP, following a Master’s in AI where I developed models for gene-based cancer treatment. My research has consistently centered on representation learning – how to model complex systems in ways that reveal their underlying structure.Before founding Thema, I was a Senior Data Scientist for the Canadian government, where I led representation learning and retrieval research from theory to deployment. I designed embedding and ranking architectures and built the production systems around them – translating advanced methods into scalable, real-world AI infrastructure.At Thema, I design embedding and unsupervised learning systems that algorithmically map market structure and adjacencies from primary data. My team develops temporal, multi-vector representations of companies across products, business models, and customer dimensions – along with AI agents for strategic intelligence.I care about one thing: turning frontier research into systems that solve hard, real-world problems.