Kamen Petroff builds AI solutions at ATOS. After years of working on real-time media streaming and web conferencing, he shifted his focus to natural language processing and deep learning in 2017. He is the creator of kpe/bert-for-tf2, an early implementation of the Transformer architecture for TensorFlow 2.
His current work explores the transition from sequence imitation to verifiable reasoning. Kamen speaks about how language models are moving past simply mimicking data, instead shifting compute to the inference phase so coding agents can continuously improve on their own. He is based in Germany.
