Platform Architecture

Four Layers. One Intelligent Twin.

Digital Duo is a full-stack AI digital twin platform. Each layer is independently powerful — together, they form a closed-loop intelligence system that sees, predicts, and acts on your physical world in real time.

Layer 01 Layer 02 Layer 03 Layer 04

System Architecture Overview

Digital Duo follows a four-tier architecture: raw physical data is ingested at the edge, normalised and streamed to the twin engine, processed by the AI inference layer, and surfaced through the visualisation and simulation interfaces.

Physical Layer

IoT sensors, SCADA, PLCs, ERP/MES APIs

Ingestion Layer

Edge agents, protocol adapters, stream normalisation

Twin Engine

State graph, synchronisation daemon, schema registry

AI Layer

Predictive models, NLQ engine, generative prescriptions

Interface Layer

Duo Dashboard, Simulation Studio, REST/GraphQL APIs

Layer 01

Data Ingestion Engine

Connect anything. Normalise everything.

The ingestion layer is the nervous system of Digital Duo. It bridges the gap between heterogeneous physical infrastructure and the unified data model that powers the twin — regardless of vendor, protocol, or data format.

Technical Specifications

Supported Protocols

MQTT, OPC-UA, Modbus TCP/RTU, BACnet, REST, GraphQL, AMQP, Kafka, JDBC

Edge Agent

Lightweight Go binary (<12 MB), deployable on ARM/x86, supports air-gapped environments

Throughput

Up to 500,000 data points/sec per ingestion node; horizontally scalable

Latency

Sub-100ms end-to-end from sensor read to twin state update

Schema Registry

Auto-infers asset schema from first 1,000 readings; supports manual override and versioning

Security

mTLS between edge agents and cloud, AES-256 at rest, SOC 2 Type II compliant

Data Flow

Sensor Read

Edge agent polls or subscribes to physical sensors at configurable intervals (1ms–1min)

Protocol Translation

Raw readings translated to canonical JSON schema via protocol-specific adapters

Stream Normalisation

Timestamps aligned to UTC, units converted, outliers flagged, gaps interpolated

Twin State Update

Normalised data written to the twin state graph via the synchronisation daemon

Native Integrations

Siemens SIMATICRockwell FactoryTalkHoneywell ForgeSAP PMIBM MaximoSalesforce Field ServiceAWS IoT CoreAzure IoT HubGoogle Cloud IoT

Layer 02

Duo Dashboard

Your physical world, rendered in real time.

The Duo Dashboard is the primary interface for operators, engineers, and executives. It renders a live 2D/3D representation of your assets, overlaid with real-time telemetry, anomaly alerts, and AI-generated insights — all in a single pane of glass.

Technical Specifications

Rendering Engine

WebGL 2.0 via Three.js; supports glTF/GLB, IFC (BIM), and STEP file imports

Asset Models

Import existing CAD/BIM models or auto-generate schematic 2D layouts from sensor topology

Telemetry Overlay

Real-time sensor values, health scores, and anomaly heatmaps rendered directly on the 3D model

Alert System

Configurable threshold alerts with severity tiers (Info / Warning / Critical); webhook + email delivery

Access Control

Role-based views: Executive (KPI summary), Engineer (full telemetry), Operator (task-focused)

API

REST and GraphQL APIs for embedding dashboard widgets into existing enterprise portals

Data Flow

Model Load

3D/2D asset model loaded from file import or auto-generated schematic

Telemetry Binding

Sensor data streams bound to model nodes via the schema registry

Real-Time Render

Dashboard re-renders at up to 60fps as new telemetry arrives from the twin engine

Alert Dispatch

Anomaly detection triggers visual alerts on the model and dispatches notifications

Native Integrations

Autodesk Revit (BIM)SolidWorksPTC CreoBentley iTwinEsri ArcGISTableauPower BIGrafana

Layer 03

Predictive Analytics Engine

Know what will happen before it does.

The core AI layer. It runs continuously against the live twin state, applying a suite of ML models trained on your asset's historical data to forecast future states, detect emerging anomalies, and generate natural-language explanations of what it finds.

Technical Specifications

Model Types

LSTM for time-series forecasting, Isolation Forest for anomaly detection, XGBoost for RUL estimation, GPT-4o for NLQ and report generation

Training Pipeline

Auto-trains on first 90 days of ingested data; retrains weekly on rolling 12-month window

Prediction Horizon

Configurable 1h–30d forecast windows; confidence intervals included on all predictions

NLQ Interface

Natural language queries against live and historical twin data; returns structured reports with cited data points

Explainability

SHAP values surfaced for every prediction; engineers can inspect which sensor readings drove each forecast

Accuracy

Median MAPE of 4.2% on RUL predictions across manufacturing pilot cohort (n=8 assets, 6-month eval)

Data Flow

Feature Extraction

Rolling statistical features (mean, std, FFT) computed from raw telemetry streams via dbt pipelines

Inference Run

LSTM and XGBoost models run inference on latest feature vectors every 60 seconds

Anomaly Scoring

Isolation Forest assigns anomaly scores; scores above threshold trigger alert pipeline

Report Generation

GPT-4o synthesises prediction outputs into human-readable reports with cited evidence

Native Integrations

OpenAI GPT-4oHugging Face Inference APIMLflow (model registry)Apache Kafka (stream processing)ClickHouse (time-series store)dbt (feature engineering)

Layer 04

What-If Simulation Studio

Test every decision before you make it.

The Simulation Studio is a sandboxed environment where engineers and planners can modify variables in the digital twin and run forward simulations to see downstream impact — without touching the physical asset. Every simulation is logged, versioned, and shareable.

Technical Specifications

Simulation Engine

Discrete-event simulation (SimPy-based) with Monte Carlo sampling for uncertainty quantification

Variable Types

Operational (speed, temperature, load), structural (layout, routing), environmental (ambient conditions)

Simulation Speed

Up to 1,000x real-time acceleration; 30-day simulation completes in under 45 seconds

Scenario Management

Unlimited saved scenarios; branch-and-compare up to 8 scenarios side-by-side

Output Metrics

OEE, MTBF, energy consumption, throughput, cost-per-unit, carbon emissions — all projected

Prescriptive AI

After each simulation, GPT-4o generates a ranked list of recommended parameter changes to optimise the target KPI

Data Flow

Scenario Setup

Engineer selects base state (current twin snapshot) and modifies target variables

Simulation Run

Discrete-event engine runs forward simulation with Monte Carlo sampling across uncertainty ranges

Result Analysis

Output metrics computed and visualised; confidence intervals shown for all projections

Prescription

GPT-4o analyses results and generates ranked recommendations with estimated ROI for each

Native Integrations

Python SimPy (simulation core)Apache Arrow (columnar data exchange)Plotly (result visualisation)Confluence / Notion (scenario export)Jira / ServiceNow (action item creation from recommendations)

Early Access

Get Deep Access to the Full Platform

Pilot partners receive full platform access, dedicated onboarding engineering, and direct input into the product roadmap. We're selecting 25 organisations across all three verticals.