Architecture & Workflow

🧩 Data Ingestion

  • Pulls from: Chainlink (on-chain), Binance/Coinbase APIs (off-chain), decentralized sentiment indexes

🔍 Preprocessing & Outlier Detection

  • Normalizes inputs into JSON

  • Filters extreme values using clustering (e.g., DBSCAN)

🤖 On-Chain Inference

  • Nodes run small ONNX models (≤50 KB) in a modified EVM

  • Each model outputs a (price, confidence) pair

🧮 Ensemble Aggregation

  • The oracle contract computes a confidence-weighted median

  • Final result stored in latestPrices[symbol]

🔁 Fee Pool

  • Consumers pay oracle usage fees

  • Subsidizes gas cost of on-chain inference

PreviousOverviewNextIntegration Examples