Unlocking On-Chain Intelligence: A Practical Framework for GenAI-Powered Smart Contracts

Abstract

Generative AI and modern Machine Learning are rapidly transforming industries, creating new tools and use cases. As these models grow, there is an increasing need for reliable Machine Learning Operations (MLOps) to manage and deploy them. At the same time, demand is rising for smarter smart contracts and more complex on-chain computation. Our paper presents a new framework that brings GenAI inference directly onto the blockchain. Using the Cosmos SDK, Ethermint, and the ONNX runtime, we enable AI models to run on-chain across multiple blockchain nodes. We evaluate the system’s feasibility, performance, and portability, showing that it can support different AI engines and model types. By enabling GenAI inference on-chain, our approach expands what smart contracts can do, especially for data-driven DeFi, and opens the door to new applications in the future.