23 systems.
Built.
something that shouldn't work.
INDRA
Real-time power quality intelligence with a 13-stage ML pipeline, built and presented to Hitachi India R&D.
Substation operators monitor 3-phase electrical systems via raw waveform data, but there is no intelligence layer between data collection and action. Faults cascade within seconds, and operators lack real-time narrative guidance.
Designed a 13-stage pipeline: waveform ingestion → windowing → feature extraction → CNN + XGBoost ensemble classification → pandapower digital twin physics validation → FAISS vector memory for historical fault search → economic and compliance scoring → AI narration. The system provides real-time operator guidance through a command-room dashboard.
Won 2nd place at VISISONICS AI'26 (judged by Hitachi India R&D). Demonstrated real-time anomaly detection, narration, and operator response flow in under 2 seconds end-to-end.