• Integrating the DeepMind and Google human data layer with external vendors (e.g., Mercor, Scale AI)

• Conducting studies to generate high-quality post-training annotation data for DeepMind models and Google ML Researchers

• Scaled Google AI Studio from prototype to 17.1 Million MAU

• Designed and built a data flywheel that improved Gemini model evaluation performance by 15%, directly influencing model quality and deployment decisions

• Created a model obfuscation and A/B testing framework used/viewed by Demis H., Koray K., and the Gemini model eval team to evaluate, iterate on, and determine which Gemini model version gets released to the public

• Led full-stack development for critical API integrations, enabling launch and adoption of flagship models including Nano Banana Pro, Imagen, and Veo

• Redesigned and shipped major UI overhauls across core product surfaces to improve usability, performance, and developer experience

• Vertex AI Studio & Google AI Studio

• Implementing Google’s newest high-speed ML/scientific computation library, JAX, in Google Cloud

• Providing a JAX deep learning virtual machine (DLVM) and container for customers

• JAX virtual machine image is up to 15 times faster than existing frameworks (PyTorch and TensorFlow)

• Publishing article discussing JAX and publishing benchmark test results against popular ML frameworks

• Automated a system dependency check to be utilized in Xray DevSecOps pipeline

• Designed standalone security pipeline to be used in all imaging modalities in GE

• Utilized Jenkins and Shell Scripting to implement pipeline

• Reduced manual processing time by 40 minutes in a 5 hour pipeline