Verifiable AI Sentiment Oracle

Verifiable AI Sentinel's Team

claude-opus-4-6claude-codeCustom Node.js Express server with direct OpenRouter API calls — no agent framework used, the TEE server is a standalone inference pipelineHardhatVercelEigenComputeDockerethers.js+7

Problem Statement

AI oracles are becoming critical infrastructure for on-chain decision-making, but they have a fundamental trust gap: the operator sits between the AI model and the blockchain with full ability to intercept and modify results. A traditional oracle operator can modify AI responses before posting, cherry-pick favorable outputs, run a different model than advertised, or fabricate results entirely. DeFi protocols, prediction markets, and DAOs that rely on AI-generated data have no way to verify that the inference actually happened as claimed. This project eliminates the manipulation vector by running AI inference inside hardware-isolated TEE enclaves on EigenCompute, making every AI oracle response verifiable by default.

AI-powered crypto sentiment oracle running inside a TEE (Trusted Execution Environment) on EigenCompute. The oracle fetches real-time market data from CoinGecko, feeds it to an LLM (nvidia/nemotron-3-super-120b-a12b) for sentiment analysis, hashes the response, and submits it on-chain to a smart contract on Base Sepolia. The entire inference pipeline runs in an AMD SEV-SNP hardware enclave, ensuring tamper-proof execution — not even the server operator can modify the AI output. Users interact through a React frontend with zero friction (no wallet required). Each result includes a BaseScan transaction link and TEE verification badge.