Models that learn from every
sensor you’ve ever shipped.
IoT-WorkS Telemetry AI ingests live sensor streams, learns per-asset baselines, and turns raw readings into anomalies, predicted failures and natural-language insights - explainable, auditable, hosted in the UK.
What does the IoT-WorkS AI Telemetry engine actually do?
Anomaly detection
Per-asset baselines learned online - isolation forest + LSTM ensemble flags drift, spikes and signature changes that fixed-threshold rules miss.
- ▸Streaming + batch detectors
- ▸Auto-baseline per asset, per shift, per season
- ▸False-positive feedback loop from operators
Predictive maintenance
Time-to-failure with calibrated confidence intervals. Models trained on your fleet’s own history, not a generic vendor dataset.
- ▸Survival models (Weibull, Cox proportional hazards)
- ▸Confidence intervals, not just point estimates
- ▸Auto-creates a job in the Booking Engine when triggered
Natural-language queries
"Which reefers in Manchester ran above -18°C for more than 30 minutes last quarter?" - answered against a structured semantic layer, not freeform RAG over PDFs.
- ▸Semantic layer over MQTT + warehouse data
- ▸Audit trail - every answer cites the rows
- ▸Slack, Teams, portal, API access
Edge inference
Run quantised models on our gateways for sub-second latency and offline operation - sites with patchy 4G keep working, only summaries hit the cloud.
- ▸ONNX-Runtime + TFLite on ARMv8 gateways
- ▸OTA model updates with rollback
- ▸GDPR-friendly: raw data can stay on site
How does data flow from a sensor to a decision?
Built on open standards. No vendor lock-in. Edge to cloud, with the option to keep raw data on-site for GDPR-sensitive deployments.
Sensors
LoRa, NB-IoT, Cellular, BLE - temp, humidity, vibration, current, GPS.
Gateways
ARMv8 + ONNX runtime. Sub-second inference, offline-first.
MQTT + REST
TLS, mTLS-capable. Buffered forwarding when network drops.
Vize ML Engine
Online learning, drift monitoring, model registry, semantic layer.
Vize Portal
Dashboards, alerts, NL query, REST + webhooks for your stack.
Integrates with what you already run
connectors as of 2026.05Where is AI telemetry already earning its keep?
Compressor failure, 7 days early
σ-shift on cycle frequency. Auto-creates a service job before product loss.
industrialBearing wear from current draw
Spectral signature of stator current - no extra vibration sensor needed.
healthcarePharmacy fridge audit-as-code
Per-fridge baseline, breach detection, MHRA-format export.
smart citiesAir-quality outliers
Detects sensor faults vs genuine pollution events with 92% precision.
energySub-meter anomaly clustering
Surfaces unusual loads against learned tenant patterns.
agricultureSoil-stress early warning
Multi-sensor fusion across moisture, temp, conductivity - 5–9 days lead.
Frequently asked questions
- What is AI Telemetry?
- AI Telemetry is IoT-WorkS's machine-learning layer that ingests live sensor streams, learns a baseline per asset, and converts raw data into anomalies, failure predictions and natural-language insights. UK-hosted, with explainability and audit trails.
- How accurate are the predictions?
- Average model accuracy across deployed estates is 96.4%. Confidence is reported per inference - a recent reefer compressor failure was flagged at 94% confidence seven days before it failed; a baseline cycling anomaly was detected at 99% confidence.
- Can AI Telemetry run on the edge without cloud connectivity?
- Yes. The platform is 100% edge-capable. Models run on a VG-E300 Edge ML Gateway with ONNX Runtime or TensorFlow Lite, sub-second inference, and a cloud sync when the network is available.
- Which industries use AI Telemetry?
- Cold-chain logistics, manufacturing (predictive maintenance), facilities and energy, agriculture and environmental monitoring. The platform is industry-agnostic - it learns from whatever sensors you connect.
- How is AI Telemetry different from a generic IoT dashboard?
- Dashboards show what is happening now. AI Telemetry models what is normal for each asset, then surfaces deviations, predicts failures, and answers natural-language questions like 'Which reefers ran above -18°C for more than 30 minutes last quarter?'.
- Is the data UK-hosted and GDPR-compliant?
- Yes. UK-hosted by default, with EU hosting at the customer's choice. GDPR-compliant. Customer data is never used to train models for other customers.
Three facts about IoT-WorkS AI Telemetry.
IoT-WorkS AI Telemetry processes more than 1,000,000 model inferences per day at 96.4% average accuracy across deployed UK and EU customer fleets, using a streaming anomaly-detection ensemble of isolation forest plus LSTM models with per-asset online baselines.
In a UK cold-chain fleet of 2,400+ refrigeration assets, IoT-WorkS AI Telemetry forecast compressor failures 7–11 days in advance with 94% confidence by detecting sigma-shifts in cycle frequency, which let the operator dispatch engineers before product loss occurred.
IoT-WorkS runs quantised ONNX-Runtime and TensorFlow Lite models on its VG-E300 ARMv8 edge gateways, delivering sub-second predictive-maintenance inference offline, so deployments on sites with patchy 4G keep operating and only summary results travel to the cloud.
Bring your own telemetry. We'll show you what's hiding in it.
Send 30 days of historical data - we'll run our anomaly + predictive engines and walk you through what we found.