AX, From Insight to Action.

Better judgement, faster execution —changing how companies and industries run.

WHAT WE DO

AX · AI Transformation

We decide where AI belongs in your work, then carry its judgement through to action.Six steps, from diagnosing the data to building and running it.

Finding AI opportunities by checking equipment and business system data

Finding the opportunity · Checking the data

We pick the work where AI pays off most, then check whether the data to train it actually exists.

Choosing candidates: Work heavy with repeated decisions and waiting, ranked by expected gain and difficulty
Data check: Judging trainability from the quality and labelling of equipment and business system data
Data collection: Equipment connected over OPC-UA, Modbus and MQTT, with operating data organised

Standardising scattered data and connecting it through shared meaning

Standardising data · Turning it into knowledge

Equipment, document and event data are organised under one shared meaning, forming a knowledge base that models and agents both draw on.

One data format: Equipment names and codes unified, with relationships defined through industry terms and an ontology
Documents and images linked: Documents, drawings and images searched by meaning, with their data relationships visible
A shared vocabulary: Time-series, events and text tied together so models and agents read the same thing

Comparing and training AI models for each operational task

Modelling · Training

We compare models against the task, security rules and operating conditions, weighing accuracy against the cost of false alarms.

Model per task: Time-series models for anomalies, maintenance and forecasting; generative AI with RAG for documents
Combining models: An in-house sLLM and a commercial LLM API combined to fit the security requirements
How we choose: Experiment tracking and cross-validation, plus what a false alarm actually costs the team

Validating AI decisions with operating data and human approval

Field validation · Human check

Field validation proves the gain, and human review and approval then set how far it runs on its own.

Measuring the effect: Before and after compared on the same basis using real operating data (PoC)
Human in the loop: The reasoning is checked and unclear cases feed the next round of training
Agreeing thresholds: Alarm levels and autonomy tuned with the team, weighing false alarms against missed ones

Connecting AI agents with business systems and equipment

AI agents · Connecting the systems

AI is wired into MES, ERP and equipment so analysis flows through to reporting, approval and action.

Into the work: Natural-language agents handle queries and reports; equipment faults follow detect → propose → approve → act → confirm
Choosing deployment: Cloud, on-premises or air-gapped — whichever the security requirements call for
System integration: Connected to MES and ERP by API so it works inside the existing flow

Monitoring model performance and repeating retraining and deployment

Operating · Keeping it learning

We keep watching how the model behaves, and when accuracy slips we retrain and redeploy.

Watching data and accuracy: Shifts in input data and model accuracy signal when to retrain
Versions and rollback: Deployment history is kept so a proven earlier version can be restored
Continuous improvement: User feedback and new data go back into training, followed by fresh validation

Applications

Where it applies

We connect data and AI to real industry work so that the way you operate actually changes.

Enterprise

AI agents automate repetitive work and support knowledge use and decisions.

Manufacturing

AI analysis of production and quality data optimises processes and automates quality.

Mobility

Vehicle and traffic data make movement and operations safer and more efficient.

Energy

Energy data and optimised demand and consumption make operations more efficient.

Public

Public data and AI automate civil, administrative and customer service work.

BUILD CASES

Selected projects

See how data became judgement, and judgement became better operations.

Autonomous paper mill operation with AI-OT

An autonomous platform joining equipment, energy, quality and safety across the paper process.

AI-OT
Predictive Maintenance
MLOps

Factory Insight AI

Early fault detection on screw-fastening and pick-and-place robots, with causes and remedies explored in conversation.

LSTM Autoencoder
LLM Assistant
LightGBM

AI Smart Optimizer

Anomaly detection and thermal mass estimation for heat-treatment equipment, with recommended process settings.

XGBoost
Transformer
Edge Gateway

See all build cases

PARTNERS & CLIENTS