Deterministic diagnostics for industrial systems
OBIS fuses auditable evidence with GPT reasoning to deliver trusted, actionable diagnostics—grounded in your data.
Deterministic Evidence
Layer 1 converts operational data into time-stamped proof stamps—no speculation.
Industrial Integrations
BACnet, Modbus, OPC UA, MQTT, historians—normalized into a single schema.
Safe AI Reasoning
Rule-bounded GPT grounded in your Layer-1 Evidence Block for structured, auditable recommendations.
How It Works
- Upload data or connect live sensors.
- Evidence Loop normalizes, baselines, and stamps anomalies.
- Suggestions grounded in evidence: root cause, actions, impact.
Layer 1 — Evidence Loop
Normalization, baselines (7/30/90 days), anomaly detection, and time-anchored proof stamps.
Layer 2 — Integration Engine
Optional real-time protocols with normalization into one schema.
Layer 3 — Suggestion Engine
GPT transforms evidence into structured outputs and actionable insights.
Integrations
BACnet, Modbus, OPC UA/DA, MQTT, Tridium Niagara, OSIsoft PI, Ignition, Wonderware, FactoryTalk, GE Proficy, AWS IoT, Azure IoT, Google IoT, AHRI/ASHRAE/OEM curves.
FAQ
How do you prevent hallucinations?
OBIS strictly grounds GPT in a deterministic Evidence Block built by Layer 1 (baseline comparisons, anomalies, Proof Stamps). We enforce rule-bounded reasoning via a master system prompt, use fact-mirroring logic and controlled innovation gates, and keep temperature low. The model cannot override the evidence.
What data do we need to start?
Start with CSV or Excel exports (or historian extracts) with timestamps and key metrics: power, temperature, pressure, vibration, load, and notes. OBIS auto-detects columns, converts units (e.g., °F/°C, psi/bar, kW/W), aligns timestamps, filters noise, and maps into a canonical schema.
Can OBIS run without live integrations?
Yes. The MVP runs entirely on file uploads. Live protocols (BACnet, Modbus, OPC UA, MQTT, Niagara, historians) are part of Layer 2 and can be added later without reworking your data.
What exactly does OBIS output
Layer 1 computes baselines (7/30/90-day windows), detects deviations (e.g., z-score/MAD thresholds, trend slopes), and matches prior anomaly signatures. It then generates time-anchored statements such as: “Compressor power increased 16.3% vs 30-day baseline.”
Which systems and protocols do you support
Protocols: BACnet (MS/TP, IP), Modbus (TCP/RTU), OPC UA/DA, MQTT; Niagara Framework. Historians: OSIsoft PI, Ignition, Wonderware, FactoryTalk, GE Proficy. Cloud IoT: AWS, Azure, and Google. Standards: AHRI/ASHRAE and OEM curves.
How do you handle security and privacy
API-key enforcement per tenant, CORS restrictions in production, and secrets stored in a professional secrets manager. Data is isolated by tenant and can be retained per policy. External audits (e.g., SSL Labs, ImmuniWeb Community) can be shared with partners.
Is Layer 2 required for the MVP
No. Layer 2 is optional and provides real-time depth. The MVP flow is: upload → Evidence Loop (Layer 1) → Suggestion Engine (Layer 3) → structured report.
How does the pattern library learn
OBIS creates signature hashes for symptom/evidence clusters. Technician feedback updates occurrence counts, success rate, and confidence for each pattern, improving future matches.
Who is OBIS for
Plant managers, reliability engineers, energy managers, and operations teams who need trusted, auditable diagnostics across equipment fleets and sites.