Engineering AI for Creativity and Curiosity on Mobile

This presentation by Bhavuk Jain discusses the process of engineering AI for creativity and curiosity on mobile devices, exploring how AI systems transition from research to scalable products.

Introduction

The talk addresses two AI products: AI wallpapers, which focus on generative creativity, and Circle to Search, a tool for contextual visual search. The discussion includes user experience (UX), latency, scalability, and safety concerns involved in bringing these AI solutions to mobile platforms.

AI Development Stages

• Post Training: Aligns the foundational model with human preferences, style, and safety guidelines.
• Fine Tuning: Specializes the model for specific use cases to enhance its effectiveness.
• Retrieval and Grounding: Ensures the system is well-grounded, reducing hallucinations by connecting to factual sources.
• Inference and Guardrails: Encompasses the deployment infrastructure and guards the model to ensure reliable and safe operations at scale.

Case Studies

1. AI Wallpapers: Allows users to generate unique wallpapers. This application faced challenges such as UX simplicity, artistic quality control, safety, and cost. The innovation was integrating AI into Android's design philosophy, thereby enhancing user interface personalization.
2. Circle to Search: Simplifies the visual search process by reducing steps. This tool leverages Google Lens' capabilities and integrates deeper with device OS for a seamless user experience, eliminating significant friction during searches.

Key Learnings

• Clear product principles and evaluation pipelines are critical for generative AI projects to ensure artistic quality.
• Deep integration with the OS can enhance user engagement by making experiences frictionless and more intuitive.
• Establishing reliable trust through grounding in factual data is essential for maintaining user trust.

Overall, the presentation highlights the importance of principled AI development, user-centric design, and the balance between freedom and guided experiences to create innovative and scalable mobile AI products.

This is the end of the AI-generated content.

This talk shares practical lessons from building production-grade AI for creativity and curiosity on mobile devices. I’ll walk through how we turned ambitious, research-driven ideas into scalable, reliable features that feel effortless to users, from generative tools for artistic expression to multimodal information experiences. Along the way, we’ll look at key decisions in model design, UX, latency, and safety, and how they come together to ship AI that genuinely delights users at scale.