We've taken a common user action – taking screenshots – and made it more useful, by making extracted and generated information from these visual captures accessible to users. Pixel Screenshots allows users to search and find information about individual captures and even synthesized information across multiple.

In a small and tight-knit team of designers and engineers we co-evolved Google's most powerful LLM - Gemini - to become the bedrock technolgy for this product. This Gemini Nano runs on-device, to make sure all information stays fully private. Beyond being responsible for the overall UX architecture, I've designed, protoyped and developed countless interaction and interface solutions to make this feel intuitive for users. I've developed text streaming that adapts to varying latencies in real time, UI layouts and worked on prompt engineering to ensure consistent information hierarchies even in generated content ... .

Below are examples of some text streaming explorations, generated in real time based on varying on-device LLM and rendering parameters. The prototype used our actual LLM models and became an integral tool for designers, UX researchers and in my communications with engineering.