Koray Kavukcuoglu built his reputation as a deep-learning researcher before he became an executive. Google’s author bio and DeepMind’s own history credit him with founding DeepMind’s deep learning team and helping pioneer two field-defining systems — DQN (deep reinforcement learning that mastered Atari from pixels — results around 2013; Nature paper 2015) and WaveNet (neural raw-audio speech synthesis later shipped across Google products).

That research spine is why his name sits at the center of Google’s AI stack today. Over roughly 13 years at DeepMind he moved from scientist to VP of Research, then CTO of Google DeepMind and Google’s Chief AI Architect — roles that wired frontier models into products — before the latest leadership change named him SVP of Google DeepMind.

This page is an evergreen career profile: technical contributions, ladder, and remit. It is not a rehash of the August 2026 org chart announcement.

Research roots: DQN and WaveNet

Pichai’s note to DeepMind staff credits Kavukcuoglu with starting the lab’s deep learning team and leading the way on WaveNet and DQN. Google’s author page repeats the same spine: VP of Research tenure, deep learning team founder, DQN and WaveNet as signature breakthroughs.

  • DQN — DeepMind’s Atari-playing Deep Q-Network showed strong results around 2013; the peer-reviewed Nature paper “Human-level control through deep reinforcement learning” followed in 2015, listing Kavukcuoglu among the authors.
  • WaveNet (2016) — a generative model of raw audio that predicted speech sample-by-sample (up to 24,000 samples per second). DeepMind says distillation made generation about 1,000× faster, producing a second of speech in roughly 50 milliseconds. WaveNet and related systems fed Google Assistant, Maps navigation, Voice Search, Cloud Text-to-Speech, and call-quality work such as WaveNetEQ.

By DeepMind’s tenth-anniversary note in 2019, Hassabis already described Kavukcuoglu as VP of Research and “a key contributor to many of DeepMind’s biggest breakthroughs, from the DQN system that mastered Atari in 2013 to the research and deployment of WaveNet.”

The climb inside DeepMind and Google

  • Deep learning lead / research scientist — early DeepMind years; founded the deep learning team.
  • VP of Research — publicly highlighted by late 2019 as DeepMind entered its second decade.
  • CTO of Google DeepMind — operational and technical leadership of the AI unit.
  • Chief AI Architect, Google — connect frontier model work with Google’s product surfaces (title he retains).
  • SVP, Google DeepMind (2026) — owns Gemini models, frontier research, and the Gemini app/developer teams; reports to Sundar Pichai.

Hassabis’s handoff note frames continuity: “The Gemini models are in good hands with Koray and the leads, as they have been for a while.”

Career ladder (at a glance)

  • Deep learning team founder / research scientist → VP of Research → CTO → Chief AI Architect → SVP, Google DeepMind.
  • Signature technical marks: DQN, WaveNet (and related speech systems in Google products).
  • Current remit: Gemini models, frontier research, Gemini app and developers; reports to Pichai; retains Chief AI Architect.

What the SVP seat owns

Under the current structure, three levers sit with Kavukcuoglu:

  • Frontier Gemini model development
  • Frontier AI research
  • The Gemini consumer app and developer tooling

That is a product-group shape — models, distribution, and builders in one reporting line to the CEO — while Hassabis remains chair and Alphabet chief scientist, advising on models and research and leaning further into Isomorphic Labs.

Why the profile matters

Google is not hiring an outside turnaround CEO for DeepMind. It is elevating the scientist who helped invent the lab’s deep-learning toolkit, shipped speech synthesis into Google products, and already sat at the research–product junction as CTO and Chief AI Architect. The bet is execution speed without discarding research DNA.

The open test for that bet sits next to Alphabet’s wider science portfolio: Hassabis on AGI strategy, and the Dean–Ghemawat Discovery Loop PBC spinning out automated scientific loops with Alphabet as founding investor and Cloud partner.

Our take

Read Kavukcuoglu as DeepMind’s bridge operator: DQN/WaveNet-era builder, research VP, then the executive asked to wire frontier models into Google’s products. The SVP title under Pichai makes that bridge the org chart — not a side role next to a lab director.

Related trilogy: DeepMind leadership reshuffle · Alphabet’s Dean–Ghemawat PBC bet.

Series: DeepMind Leadership — reshuffle, Kavukcuoglu profile, and Discovery Loop PBC analysis.

Sources