Predictive maintenance, from nothing, on a mountain.

The Riesenrutschbahn at Poppeltal runs fifty bobs down a mountain every operating day. Before this engagement, none had any condition monitoring — maintenance was time-based: inspect on a schedule, replace on a calendar, hope nothing fails in between. No visibility into vibration, bearing wear, or load on any individual bob. The site connectivity matched: staff WiFi at the base, a remote staff cabin with no internet at all, no IoT infrastructure. Nothing. Apuna built both.

Each of the fifty bobs now carries an autosen vibration sensor sampling at 1 kHz on a 900 mAh battery — vibration only, reporting directly over LTE Cat-M, no on-site gateway, no proprietary cloud. During commissioning the team evaluated LoRa for the link (range-testing with two Heltec units across the terrain) and made the honest call: LTE Cat-M direct was the right radio for this mountain — no LoRa infrastructure deployed. Each bob also carries an RFID tag; an IdentControl read head at the lift hill registers every pass, producing a per-bob usage count.

A Raspberry Pi 4 CM4 in an industrial enclosure — on an isolated IoT VLAN — runs the Mosquitto broker and the IdentControl integration, forwarding both streams to Cloudflare: Workers → Queues (dead-letter queue; no reading lost) → D1 for history, plus a Durable Object per bob for live state.

Fusing vibration and usage, a Worker runs ISO 10816 RMS-velocity severity, spectral and bearing-defect features (crest factor, kurtosis), and per-bob anomaly baselines normalised against usage. Explainable, no fault-labelled history required to start, and it tells staff which bob is behaving unusually before it tells them what to do. A human decides. Not a language model — the right tool for vibration time-series: standards-based, defensible, no hallucinations. As labelled events accumulate, the platform is designed to graduate to compact supervised models via Workers AI for remaining-useful-life. That is roadmap; what is live is already useful. Staff read it on a mobile-optimised Grafana dashboard, self-hosted on Cloudflare Containers, secured via Access. The phone in their pocket is the dashboard.

Three VLANs keep operations clean — Staff, Admin, IoT. Visitors get an open Freifunk mesh, no password, no access to operations; the two worlds do not touch. And the remote staff cabin — two people, no internet, ever — now has Starlink. Not an afterthought; the obvious thing to do when the gap exists and the technology to close it does too.

Hardware and expenses pass through at cost. The build was contracted on a Statement of Work. The ongoing relationship runs on an Apuna Care subscription — Apuna's first recurring revenue. Not a one-time project, not hourly consulting: a customer paying for continuity, and Apuna delivering it.

Operations first. AI last, riding on clean data, not substituting for it. Open components. Data on the customer's own estate, sovereign and portable. A human always decides. And Care means the engagement does not end when the build does.