Technology

ANNA - automated AI modelling and operational optimisation

ANNA (Advanced Neural Networks and AI) is UnifAI Technology's AI modelling and optimisation engine. It automates the development of predictive models and the decision rules that make their outputs useful in real-world operations. Models are trained, tested and refined using the application's data, with performance assessed against its objectives, constraints and trade-offs - not accuracy alone.

ANNA is model-agnostic: it is not a single AI model or a fixed prediction algorithm. It trains and compares different modelling approaches, then optimises the model and the rules used to act on its predictions.

Predictive models with an operational purpose

ANNA can support forecasting, virtual sensors, anomaly detection and decision support across different sectors. The same framework can develop models for different sites, assets and tasks, using the data relevant to each application.

Predictions and response rules can feed dashboards, alerts, APIs and existing operational workflows.

You define the outcome. ANNA runs the modelling process.

Your team defines the outcome and constraints. ANNA automatically runs the preparation, comparison and refinement around them, testing which inputs, model settings and response rules best serve the objective.

Each round guides the next combinations to test. ANNA then selects and validates the model and operating rules against the agreed criteria before use.

Your operational briefYou define the outcome and constraints.ANNA automates the modelling and optimisation around them.
  1. ANNA prepares and tests the inputs

    ANNA cleans and structures the data, and tests potentially useful inputs such as trends, rates of change and relationships between measurements.

  2. ANNA trains and compares models

    ANNA trains and compares different AI models - alternative ways of making the prediction - and tests which approaches and settings best serve the agreed objective.

  3. ANNA optimises the model and the response

    ANNA refines promising model configurations and tests decision thresholds and operating rules against the practical costs, risks and constraints of the application.

  4. ANNA validates before use

    ANNA validates the selected model and operating rules against the agreed criteria, and records the configuration and results for review.

Each trial guides the next combination of inputs, model settings and response rules.

Models that evolve with your data

ANNA is being developed to continue this cycle after deployment: monitor the model, retrain and re-optimise using new evidence, and revalidate before any replacement is introduced. A newly trained model replaces the live model only when it meets the agreed criteria and demonstrates an improvement. Otherwise, the existing model is retained, or flagged for review if it no longer meets those criteria.

Scheduled checks, as well as new evidence, trigger review of data availability, operating conditions and model performance. The aim is not a one-off model build. It is a controlled cycle of improvement.

Development direction - post-deployment lifecycle
A controlled cycle of improvement
  1. Connect dataEstablish the inputs
  2. Test inputsFind useful information
  3. Train and optimiseCompare model and rule configurations
  4. ValidateCheck the agreed criteria
  5. Deploy or retainIntroduce a validated improvement
  6. MonitorReview performance and evidence

Monitoring feeds the next round of input testing, training and validation.

Scale the modelling process, not one universal model

A different site can need a different model without requiring the development process to be built again.

What is reused is the development framework. The models themselves are automatically trained and evaluated using the relevant data and objectives for each site, asset or task.

This is how ANNA is designed to scale: by reducing the repeated specialist modelling work needed to extend AI across an estate, while retaining evaluation against each application's objectives.

Explainable decisions and auditable development

ANNA retains structured records of the models, configurations, decision thresholds, constraints and evaluation results it tests. Simulations test how alternative response rules change performance, cost and risk.

These records support review of why a configuration was selected and how a prediction will be used. Where the intended use requires additional model-level explanation methods, those requirements can be built into the application design.

Working with your team

Your team brings the operational knowledge; ANNA automates the modelling and optimisation. We configure the application with your team and integrate the outputs into your operational systems. DAVE can deliver data and model outputs to your existing platforms and dashboards without requiring its visualisation layer.

Our longer-term aim is to put this workflow directly into the hands of domain specialists, so they can define the outcome and constraints without building the underlying modelling pipeline themselves.

Name
ANNA - Advanced Neural Networks and AI
Role
Automated modelling and operational optimisation
Delivery
Dashboards, alerts, APIs and operational workflows
Application evidence

Predictive modelling in a real application

At Warleigh Weir, AI predictions were compared with laboratory E. coli results and used within a public water-quality information workflow.

Explore the Warleigh Weir case study ↗

Warleigh Weir: laboratory E. coli results (blue) and AI predictions (orange). The case study records R² = 0.869.

Configure ANNA around your objective

Start with what you need to predict, what a useful response looks like and the constraints the system must respect.

Discuss ANNA

Image detail

Privacy preferences

This website uses necessary technology to deliver pages and handle enquiries. Optional analytics, advertising and video embeds are switched off.

Wix hosting and security — necessary
Wix delivers the website and protects requests. The sec-fetch-unsupported security cookie lasts for the browser session.
Wix enquiry session — necessary
The wixSession cookie supports the visitor session used by the enquiry form. It lasts for the browser session.
Analytics, advertising and embedded video — optional
Off. No optional tracking tags, advertising cookies or video players are loaded by this website. External links open the provider’s own service.

See the Cookie Notice and Privacy Notice.