Join DeepMind India's Agents team in Bengaluru, dedicated to advancing foundational capabilities in Large Models to achieve Artificial General Intelligence (AGI).

Our mission involves exploring emergent agentic behaviors using novel reinforcement learning algorithms at scale. We aim to translate research breakthroughs into impactful applications for Gemini and critical Google product launches, reaching hundreds of millions of users.

This role offers a unique opportunity to contribute to the forefront of AI research, focusing on developing more capable autonomous agents through reinforcement learning and ML optimization methods. You will play a key role in designing novel algorithmic architectures with the ultimate goal of solving and building AGI.

Design, implement, and rigorously evaluate models, agents, and software prototypes for large foundational models.

Push the boundaries of reinforcement learning and ML optimization techniques to develop sophisticated autonomous agents.

Clearly and efficiently report and present research findings, developments, and results both internally and externally through written and verbal communication.

Propose and actively participate in collaborations within the team and with external research labs, fostering strong relationships.

Collaborate closely with our Responsible AI teams to ensure ethical development of advanced intelligence, maximizing benefits for humanity.

A PhD degree in Computer Science, Artificial Intelligence, or a related field is required, or equivalent practical experience.

Significant experience, at least 8 years, with machine learning frameworks such as JAX, TensorFlow, or PyTorch is essential.

Demonstrated experience in advanced deep learning and reinforcement learning techniques is a must.

An established publication record in reputable machine learning conferences or journals like NeurIPS, ICML, ICLR, KDD, or AAAI is expected.

Preferred qualifications include experience with multimodal learning, large language models, or assistive AI agents, as well as prompt engineering, few-shot learning, post-training techniques, and evaluations.