Industry scenarios
Worked examples, each starting from a real problem an industry needs solved and showing how to solve it with the SDK. They're organised by sector — pick yours and work through its scenarios, which build on each other across the value chain.
Every scenario is self-contained: a Set up demo data block (run once) → the
solution → See the result. Plan on ~5 minutes to read one, or ~10–15 minutes
to run it end to end from an empty workspace (a couple of minutes more the very first
time, to install the SDK and set BASE_URL/TOKEN). When you're ready to go past CRUD
into prediction, the Advanced scenarios (AI) — no
data-science background needed — run ~20 minutes to 2 hours each.
Oil & gas
Upstream to downstream — the full value chain.
- Drilling operations — catch a well-control kick in its first seconds
- Production monitoring — ESP anomaly detection on a field graph
- Pipeline integrity — detect and locate a leak on the network
- Refinery operations — keep units on-spec and energy-efficient
- Tank storage & custody — reconcile inventory, catch unexplained loss
- Emissions & flaring — methane & flaring compliance reporting
Energy & utilities
- Grid telemetry — SCADA roll-ups and overload events
- Wind generation — capacity factor for a wind farm
- EV charging network — uptime and grid load
- Water distribution — trace a contamination source upstream
- Waste collection — collect on fill level, not a calendar
- Air quality — heterogeneous sensors, correct units
Manufacturing & process
- Discrete manufacturing — OEE from station telemetry
- Semiconductor — trace a yield excursion to one tool
- Pharma — exact batch records and upstream genealogy
Transport & logistics
- Rail — is the line still connected after a closure?
- Maritime — a reefer cold-chain alarm mid-ocean
- Airports — catch a turnaround slipping before the slot is lost
- Ports — crane productivity and yard congestion
- Last-mile delivery — hit delivery windows
- Retail replenishment — per-region demand, segmented by dataset
- Aerospace fleet — a suspect part's blast radius across the fleet
- Supply chain — trace a recall downstream from a bad lot
Financial services
- Portfolio & risk — exact-decimal metrics and risk-limit events
- AML & payments — expand a flagged payment into its ring
- Insurance fraud — expand a claim into its fraud ring
Healthcare
- Hospital operations — a live capacity wall-board before a ward overflows
- Medical-device fleet — device availability and fast recall tracing
- Patient flow — find the bottleneck stage in the patient journey
- Pharmacy cold chain — protect temperature-sensitive stock
Mining & metals
- Mining operations — equipment uptime and true throughput against plan
- Processing plant — protect metal recovery at the mill
- Tailings dam safety — trigger-based monitoring of a high-consequence dam
Agriculture & food
- Precision farming — drone imagery alongside soil sensors
- Salmon farming — catch an oxygen crash in time to aerate
Technology & operations
- Observability (IT ops) — alert on an SLO breach the instant it happens
- Network performance — localize degradation to shared backhaul
- Data centers — catch a thermal problem before racks throttle
Built environment
- Construction — drawings and permits tied to the site
- Smart buildings — flag a stuffy zone before the complaints
Going deeper
The Advanced scenarios (AI) take these further into real algorithms — forecasting, classification, clustering and anomaly detection. No data-science background needed: start with Machine learning, gently for the ideas in plain language, then Generate sample data for a sandbox to run them against. To clean and feature-engineer data while keeping every step traceable, see Data pipeline & lineage. The models below are grouped by what they do.
Detect & monitor condition
Learn what healthy looks like, then surface the drift.
- Predictive maintenance — flag a machine drifting from its healthy vibration signature (~1–2 hr)
- Asset health scoring — one composite health index from many signals (~20–30 min)
- LSTM anomaly detection — catch a multivariate event from the joint pattern (~2 hr)
- Process monitoring with PCA — spot a break in many sensors' normal relationship (~1–1.5 hr)
Forecast & predict ahead
Predict what's coming while there's still time to act.
- Demand forecasting — forecast load or sales with gradient boosting (~1 hr)
- LSTM forecasting — forecast a decline curve with a neural network (~2 hr)
- Early warning — predict a salmon-pen oxygen crash before it happens (~1 hr)
- Failure prediction with XGBoost — predict an ESP failure a week out, with feature importance (~1–2 hr)
Classify, cluster & infer
Label, group or estimate from features.
- Fraud classification — graph features + a model to rank fraud/AML alerts (~1–2 hr)
- K-Means clustering — discover operating regimes, asset cohorts and network communities (~1–1.5 hr)
- Soft sensor with Random Forest — infer a lab quality from cheap sensors (~1 hr)