Paige Bailey
Conference affiliation: Google DeepMind · 2026
Related work and projects
- openstax-intro-stats ↗github.com · Source ↗
- signals-and-systems ↗github.com · Source ↗
- thinking-in-data ↗github.com · Source ↗
Paige Bailey is an engineering lead for developer relations at Google DeepMind and a contributor to the research behind PaLM 2, Gemini, and Gemma. Her career spans geophysics, scientific Python, machine-learning infrastructure, developer tools, and frontier models, with a consistent emphasis on making advanced capabilities accessible to working developers.
Bailey began working with machine learning around 2009, contributing to open-source scientific-computing projects including NumPy, SciPy, Matplotlib, and scikit-learn. She studied geophysics and applied mathematics, pursued graduate work in computer science and carbonate geology, and worked at Chevron on subsurface geoscience, velocity modeling, drilling optimization, distributed computing, and GPU-based workloads.
She contributed to TensorFlow, worked on machine-learning developer experiences at Microsoft and Google, and joined GitHub, where her work included VS Code and early GitHub Copilot user-experience testing. Returning to Google, she contributed to the PaLM 2, Gemini, and Gemma model programs before leading developer-relations engineering at Google DeepMind.
Her independent projects include thinking-in-data, a VS Code extension pack for exploring and visualizing data; signals-and-systems, interactive visualizations for an open-source engineering textbook; and Gemini and Gemma examples, practical notebooks for contemporary AI models.
- Open models as practical independence. Bailey advocates models developers can download, adapt, and run on their own infrastructure, including offline or privacy-sensitive environments. Her characteristically irreverent case for open models emphasizes affordability, stability, and freedom from unexpected platform changes.
- Model selection as an engineering decision. She evaluates models against cost, latency, tools, and actual task performance. Her hands-on Gemini demonstration shows a smaller model using sandboxed Python to analyze images quickly; her assessment of computer-use benchmarks for CAD argues that affordable systems approaching frontier performance could broaden access to physical-design tools.
- Multimodal applications with persistent state. Bailey turns model capabilities into complete software: one example identifies books from a photograph, supplements missing details through search, authenticates users, and stores their libraries in Firebase. She also distinguishes Project Genie’s dynamically generated video environments from exportable three-dimensional game assets.
- Generative media with creative control and safeguards. Her work with Veo addresses camera control, visual consistency, synchronized dialogue, and SynthID watermarking. A commercial-reconstruction example contrasts manually stitching together generated video, speech, and music with a more integrated Veo 3 workflow.
- Embodied AI with separated control layers. In robotics, Bailey distinguishes multimodal perception and high-level planning from specialized software executing physical movement locally—a practical architecture for applying advanced models to hardware without confusing reasoning with motor control.
Read the topics behind these talks
5 conference talks
▶ Watch ↗AI Engineer World's Fair 202610:53
Research to Reality with Google DeepMind — Paige Bailey, Google DeepMind
Explore the talk →Loading more talks and highlights…
References
- github.com/dynamicwebpaige/openstax-intro-stats ↗
- github.com/dynamicwebpaige ↗
- github.com/dynamicwebpaige/signals-and-systems ↗
- github.com/dynamicwebpaige/thinking-in-data ↗
- ai.engineer/talks/build-deploy-ai-powered-apps ↗
- ai.engineer/talks/build-deploy-ai-powered-apps-18cc1b ↗
- ai.engineer/talks/veo-3-for-developers ↗
- ai.engineer/europe/2026 ↗
- ai.engineer/worldsfair/2026 ↗
- ai.engineer/worldsfair/2025 ↗



