Engineering team

The engine for creating AI breakthroughs, faster

Overview

Our engineers help accelerate research by building usable research tools, scaling up algorithms, and creating challenging virtual and physical worlds for testing our systems. 

Conducting world-class research requires solving difficult engineering problems. Tools and infrastructure developed by our engineering teams have successfully enabled our most significant research milestones, such as AlphaGo, AlphaStar, and AlphaFold.

Building for scale

These systems enable training of large-scale neural networks by unlocking scalable, parallel computation across diverse hardware. Internal tools for research empower our research team to run experiments seamlessly and make rapid scientific progress at scale.

Our multidisciplinary engineering team, with expertise ranging from software, hardware, and research engineers to designers, artists, and program managers, work across all DeepMind teams to deliver high-impact, state-of-the-art research. Many of our tools, libraries, environments, and papers are available open source. Explore the team’s work on our GitHub page.

Engineering highlights

Accelerating research with tools, platforms, and environments

TF-Replicator: Distributed machine learning for researchers

A software library to help deploy TensorFlow models on GPUs and Cloud TPUs with minimal effort.

TRFL: A library of reinforcement learning building blocks

An open source library of building blocks for writing reinforcement learning (RL) agents in TensorFlow.

DeepMind Control Suite: Benchmarks for RL agents

A set of continuous control tasks with a standardised structure and interpretable rewards.

Engineering teams

Engineering in Research

Research engineers and software engineers on the Research team tackle unique engineering challenges that combine state-of-the-art computer systems and AI algorithms. This is done by developing prototypes and tools that allow our teams to perform rigorous experimentation at scale. This includes creating complex reinforcement learning agents and training pipelines alongside tools for visualisation, debugging, testing, and running reliable agents.