LLM-Integrated Hybrid Power Monitor & Source Switch

Small-scale solar and battery hybrid installations frequently experience poor energy utilization due to inefficient manual source changeover and opaque inverter telemetry. Most commercial monitoring tools simply present dense tabular numbers that non-specialist users struggle to translate into actionable maintenance or energy conservation actions.

System Designer and Developer. Designed the dual-source electrical sensing circuitry, programmed the microcontroller source-switching algorithm, and developed the Python pipeline for local LLM telemetry ingestion and diagnostic generation.

An Arduino sensing module samples voltage and current from both solar photovoltaic and AC grid supplies. An automated relay switching circuit executes source changeovers, prioritizing solar generation whenever insolation and battery voltage satisfy defined thresholds. Serial telemetry is streamed into a local computing node where a lightweight on-device LLM processes sliding-window time-series data to detect abnormal consumption patterns and generate clear diagnostic explanations.

Formulated a hysteresis-based switching algorithm in embedded C++ to eliminate rapid relay cycling during transient cloud cover. Built a Python telemetry service that converts electrical metrics into structured context prompts. Guided the local language model to identify load imbalances, calculate estimated solar savings, and flag abnormal voltage sags without requiring external internet access or proprietary cloud APIs.

Verified seamless automatic source transfer with minimal load disturbance. Confirmed that local LLM diagnostics accurately flagged simulated voltage anomalies and generated actionable, human-readable operational summaries.