Built RECON (Renewable Energy Certificate Observation Network) to eliminate double-counting, false telemetry, and fraudulent issuance in clean energy markets using an end-to-end ML verification pipeline paired with blockchain registries:
-
Multi-Stage AI/ML Fraud Pipeline:
• Deep Autoencoders: Reconstruct normal time-series generation profiles from power meters/IoT telemetry; elevated reconstruction error isolates anomalous spikes and unfeasible power yields.
• Isolation Forests: Unsupervised partition-based anomaly scoring to instantly isolate multidimensional outliers across solar irradiance, wind speed, capacity factors, and reported MWh.
• XGBoost Risk Classifier: Gradient-boosted decision trees trained on labeled behavioral signatures to aggregate features into a calibrated composite fraud risk score (0–100%). -
Blockchain Registry (ERC-721):
• Built and deployed smart contracts on Ethereum Sepolia.
• Automatically blocks minting if ML risk scores exceed threshold limits.
• Issues immutable certificate tokens upon verified generation and burns/retires tokens upon market redemption to prevent double-claiming. -
Full-Stack Telemetry Dashboard:
• Built an analytics dashboard displaying real-time generation metrics, sensor deviations, anomaly scores, and on-chain verification transaction proofs.
