Abstract
Large Language Models have revolutionized AI — but their centralized nature comes with real costs: concentration of knowledge, privacy risks, and latency. What if Kotlin developers could flip the script, bringing powerful AI directly onto devices with an open-source, developer-friendly machine learning framework? In this talk, we'll take you inside our journey of building SKaiNET, a brand-new framework written from scratch in Kotlin Multiplatform. We'll start by exploring the limitations of today's tools, then walk step-by-step through defining neural networks in Kotlin using a type-safe DSL, compiling them into compute graphs, and running lightweight models like convolutional nets and compact LLMs completely offline — all in pure Kotlin. You'll learn how Kotlin's strengths — coroutines, type safety, and multiplatform support — enable low-latency, privacy-first AI that runs directly on-device. With features like flexible tensor layouts and support for multiple file formats, SKaiNET bridges the gap between mobile development and data science. It's open-source. It's powerful. And you're invited to help shape its future. Expect code, insights, and a hands-on look at how Kotlin (and SKaiNET) is quietly redefining mobile AI.