Abstract
When adding artificial intelligence (AI) features to devices, on-device models are deployed to enable frictionless access to operating system information. However, building and deploying production-ready on-device models takes large investments. The end quality and usable feature set is often not understood until months into the development cycle. This disclosure describes techniques that enable rapid prototyping and validation of artificial intelligence (AI) and non-AI features in smartphones prior to investing in production-ready on-device AI models. Layers of AI agents and flexible client APIs enable access to rich data that correlates with end-user experience, resulting in substantial gains in productivity and improved product quality and user experience. Interactions with various surfaces can be handled directly via large language model (LLM) prompts, enabling rapid iteration of new features on and across surfaces. A prototyping platform for AI experiences as described herein provides AI model developers a quick and easy way to explore how context affects the accuracy and timeliness of AI-generated suggestions.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Morrino, Michael Digman; Bernal, Marvin; and Geilfuss, Bradley, "Rapid Prototyping and User Research Using On-device Context and Tooling", Technical Disclosure Commons, (February 05, 2026)
https://www.tdcommons.org/dpubs_series/9282