AI Meter Reading and Anomaly Detection System

Technicians photograph meters, AI reads them, and MeterLog flags leaks, fuel loss and missed rounds the same day, not when the bill arrives.

Problem Facilities teams still track water, power, gas and diesel by hand: numbers scribbled on clipboards or sent as WhatsApp photos that nobody compares with yesterday’s. A running leak or a draining fuel tank goes unnoticed for weeks, until the utility bill or an empty generator exposes it. By then the money is gone, and there’s no record of when it started.

MeterLog turns daily meter rounds into an early-warning system. A technician snaps a photo of any meter or gauge, AI reads the value, and the technician confirms it in seconds. Every reading is checked on the server against that meter’s own history, so unusual consumption becomes a ticket with photo evidence, a chart and likely causes, assigned to someone the same day. Managers get one dashboard for every site, email alerts for high-severity issues, and a plain-English briefing every Monday.

Two roles with data-level security: technicians can only see their assigned sites, enforced by entity access rules. All technician writes go through backend functions that check access on the server.

AI meter reading: photo upload plus an InvokeLLM vision call returning structured JSON (value, confidence, issues). The AI result is stored on the server and can’t be altered by the device, so the 92% AI read accuracy shown on the dashboard is trustworthy.

Server-side anomaly detection: a backend function on an entity automation applies five rules (backwards meter, spike, out of range, low level, and fuel loss checked against generator run-hours), with deduplication. A daily automation flags missed readings.

Anomaly tickets: side-by-side photo evidence, a 14-day chart with the normal band, AI-written likely causes, assignment, status workflow, a full activity log, and a one-click “mark as misread” that corrects the history.

Charts per asset type: daily-use bars against the expected band for meters, and level lines with refills marked for fuel tanks, overlaid with generator run-hours.

Automations and AI briefings: email alerts to managers on high-severity anomalies, and a scheduled weekly AI briefing with three concrete actions per site.

Ready to explore: a public read-only demo with three sites, 14 meters and 30 days of data. It includes a planted water leak (4.4× normal), more than 180 L of diesel lost while the generator barely ran, a corrected misread, and a missed round.

Built with: Base44 entities and access rules, 11 backend functions, 3 automations, InvokeLLM vision, UploadFile and SendEmail.

Links: meterlog.base44.app · meterlog.base44.app/demo