I am a Director & Research Scientist in the Google DeepMind team in Paris.

Between 2019 and 2022, I led fundamental research on distributional RL, imitation learning, offline RL, and RL from human feedback. I am very proud to be a contributor of the Acme library.

In late 22/early 23, I built the RLHF layer of the first version of Bard (now "Gemini App").

Between 2023 and 2025, I led post-training of Gemma models which includes Gemma 1, Gemma 2, Gemma 3. I also led the post-training of Gemini Nano  3.

Since early 2026, I lead pre-training of Gemma & pre-training for Gemini first-party models for low latency / high throughput use cases such as Search, Ads and YouTube.

contact: robert (dot) dadashi (at) gmail (dot) com