GridSentry

Distribution companies can lose a significant share of energy pushed through low-voltage networks. GridSentry asks a simple question: what if the network could report suspicious losses live instead of discovering them later?

What it does

GridSentry is a smart-grid control-room prototype for monitoring 18 smart meters across two feeders. It combines live telemetry, deterministic detection rules, and AI-assisted investigation reports to help identify:

• Energy theft and suspicious consumption patterns
• Electrical faults
• Meter tampering
• Transformer loss and feeder accounting issues
• Fleet health problems

The system exposes a token-authenticated /api/ingest/{meter_id} endpoint and includes an ESP32 + CT-clamp wiring path for physical telemetry ingestion.

How it works

The Python/FastAPI backend receives or simulates meter telemetry, stores it with SQLAlchemy, and applies deterministic rules to detect anomalies. The React/TypeScript/Vite frontend presents the results as a control-room console. For investigation summaries, GridSentry can use Groq's openai/gpt-oss-20b model to generate grounded reports from the evidence produced by the detection engine, with a deterministic fallback when AI is unavailable.

Built with

Python, FastAPI, SQLAlchemy, SQLite/PostgreSQL (Neon), React, TypeScript, Vite, Groq (openai/gpt-oss-20b), Playwright, pytest, GitHub Actions, Docker, ESP32 + CT-clamp.

Quality and testing

The project includes 15 pytest tests, GitHub Actions CI, Docker support, and Playwright-based recording/QA.

Why it is different

GridSentry combines practical smart-meter telemetry, deterministic anomaly detection, and an AI investigator in one focused control-room workflow. The core detection logic remains explainable and deterministic; AI is used to help investigators interpret the evidence rather than replace the underlying detection system.

Challenges

The main challenge was making the system useful as both a realistic smart-grid prototype and an understandable hackathon demonstration. The project therefore focuses on a clear end-to-end flow: telemetry → detection → evidence → investigation report.

Next steps

Future work could include more physical meter integrations, richer feeder-level analytics, historical trend analysis, additional anomaly models, and deployment to utility-scale infrastructure.

AI disclosure

AI was used for the investigator-report layer. Groq's openai/gpt-oss-20b generates reports grounded in evidence supplied by the deterministic detection engine. The detection logic itself is deterministic.

Built With

• ct-clamp
• docker
• esp32
• fastapi
• github-actions
• gpt-oss-20b
• groq
• neon
• openai
• playwright
• postgresql
• pytest
• python
• react
• sqlalchemy
• sqlite
• typescript
• vite