Member of Technical Staff at Google DeepMind, open-source developer and AI educator. He created BERTopic and KeyBERT and co-teaches visual explanations of language models.[1][2][6]
Start with BERTopic →
X · MaartenGr Career
Grootendorst’s academic background includes master’s degrees in organizational psychology, clinical psychology and data science and entrepreneurship. His JADS training connected that background with applied data science. At Google DeepMind he works on developer experience around Gemma and Gemini. His teaching and open-source tools make model behavior and text analysis easier to inspect.[1][5][4]
Contributions
Separate topic discovery from topic naming
BERTopic embeds documents, reduces their dimensionality and clusters them, then uses class-based TF-IDF to identify words that distinguish each group. Those steps are modular: developers can replace the embedding, clustering or representation layer without replacing the whole pipeline. The result connects semantic grouping with interpretable topic labels and visualizations.[2]
Rank candidate phrases by meaning
KeyBERT embeds a document and its candidate words or phrases, then ranks their cosine similarity. Optional diversity methods reduce repeated near-equivalent phrases in the result. This provides a small, inspectable keyword-extraction pipeline that can be adapted with different embedding models.[6]
Tools, writing & talks
Tools
Start here
BERTopic
github.comA modular topic-modeling package connecting semantic document clusters with class-based TF-IDF topic representations.
Start with BERTopic’s minimal example and inspect the representative documents behind a topic label. Python and basic embeddings are useful prerequisites; change one pipeline component at a time.
Creator[2]
Writing, research & talks
A visual course taught with Jay Alammar.
Co-instructor[7]Original source