DeRAG-Oracle: A Decentralized Retrieval-Augmented Generation Architecture for On-Chain Risk Telemetry

Smart contracts on the Ethereum Virtual Machine (EVM) cannot natively access unstructured enterprise data due to prohibitive on-chain storage costs and the lack of native text-parsing capabilities. Existing oracles remain bottlenecked by their reliance on strictly deterministic, structured data feeds. We propose DeRAG, a decentralized oracle architecture combining BNB Greenfield object storage, a FastAPI middleware powered by the Google Gemini Large Language Model (LLM), and Chainlink's Custom Runtime Environment (CRE). DeRAG distills complex, unstructured enterprise data into ultra-compact structured payloads that provide a deterministic representation of the probabilistic underlying inference, which can then be securely pushed to on-chain smart contracts. Evaluated over N=30 iterations, the architecture achieved a 100.0% execution success rate with a mean execution latency of 6.882 seconds, reducing raw enterprise data into average payloads of 369.43 bytes. By shifting heavy computation and Retrieval-Augmented Generation (RAG) off-chain, we demonstrate a robust bridge between enterprise data and Web3, enabling stable intelligence delivery designed to minimize on-chain data footprint.