See the missing AI Data Layer in action
Scroll through the seven steps that turn raw PLC data into AI-ready information — and the Solve loop that turns it into an autonomous decision.
Connect
Read tags read-only from any PLC (Siemens, Rockwell, Beckhoff) and push them over outbound MQTT/TLS to the broker. No inbound firewall changes, no vendor lock-in.
OPC-UA · MQTT · MonsterMQ
Condition
Normalise units (psi → bar, °F → °C), deadband-filter, deduplicate retries, drop out-of-range. A clean, trustworthy stream before the rest of the stack touches it.
MonsterMQ flow engine · plausibility checks
Model — the heart of the stack
Wrap every raw value with ISA-95 context: enterprise → site → area → equipment → tag, plus unit, range and status. TIC-001 = 67.3 becomes 'Pasteurizer · 67.3°C · Normal · OK'. This is the layer AVEVA Connect charges €40K/year for.
ISA-95 ontology · JSON · Neo4j mirror
Store
Write to MongoDB time-series, write-time enriched with order_id, sku, phase. Every datapoint is queryable per recipe and per production order without runtime joins.
MongoDB 7 · time-series · 90d retention
Orchestrate
N8N workflows listen for events, aggregate windows, compare against recipe-specific thresholds, and trigger the decision. No polling, no manual steps.
N8N · event-driven · webhook-out
Visualize
Grafana for analysts, Next.js for operators. Both query through FastAPI — no direct database access. Dashboards show information, not just numbers.
Grafana · Next.js · FastAPI
Distribute
FastAPI exposes endpoints, an edge subscriber writes the override back to the PLC over OPC-UA. The loop closes: from sensor → decision → action on the plant floor.
FastAPI · retained MQTT · OPC-UA write-back
Solve is not an 8th step. Solve is the test.
The 7 steps tell you HOW to build the layer. Solve — raw data triggering a real autonomous decision inside the time window where it still matters — tells you whether the build WORKS.
If the architecture doesn't flow all the way through the 7 steps to Solve, the layer stays 'trapped in the middle'. Pretty dashboards, no ROI. You can build all 7 steps perfectly and still fail — if you don't push through to Solve.
Ready to build the layer yourself?
Download the free 7-step framework or read the full case studies — beverage bottling, dairy, specialty chemicals, automotive.