Predict failures before they happen
LSTM anomaly detection on industrial sensor streams, a semantic knowledge graph of equipment failure modes, and four LangGraph agents that trace root causes and explain exactly what went wrong — not just that something did.
Free while in beta · No credit card
Four components. One pipeline.
Each piece was built separately, validated separately, then wired together. Here's what each one actually does.
Anomaly Detection
87.3% F1LSTM autoencoder trained on industrial sensor data detects deviations in real time — air temp, torque, RPM, tool wear and more.
Multi-Agent RCA
84.6% accuracy4 specialized LangGraph agents — Diagnostic, Reasoning, Planning, and Learning — collaborate to trace the root cause of every fault.
Knowledge Graph
50+ entitiesA semantic ontology of 50+ equipment entities and SWRL rules enables contextual reasoning far beyond simple threshold alerts.
Predictive Analytics
Ensemble modelEnsemble scoring fuses LSTM and Random Forest signals to give an equipment health score that decays in real time as anomalies accumulate.
The pipeline, end to end
Sensor reading in, root cause report out. These are the four stages, in order.
Ingest Sensor Data
Push air temperature, RPM, torque, tool wear and 9 other sensor readings via REST API or live stream.
Detect Anomalies
The LSTM autoencoder computes reconstruction error in under 2 seconds and raises an alert when thresholds are crossed.
Trace Root Cause
LangGraph agents query the knowledge graph, reason over failure patterns, and return a ranked list of probable root causes.
Act & Prevent
Receive prioritised maintenance recommendations and push tasks directly to your CMMS or team dashboard.
Tested on real industrial data
No synthetic benchmarks. Both datasets are publicly available — you can replicate every number.
AI4I 2020
10,000 production records · 5 failure modes · UCI ML Repository
MetroPT-3
Air compressor telemetry · Porto Metro fleet · Cross-domain transfer
It's a working system, not a demo.
The API is live. The models are trained on real data. Send a sensor reading and get a root cause report back — no setup required.