Vivaldi Digital Solutions
← Back to all solutions
vKI StudioYour own AI modelsNew

Train your own AI models: directly within your company.

vKI Studio turns your production data into a controlled learning cycle and enables customers to further develop AI models close to the real process, without giving up approval, versioning and quality assurance.

From inspection operations to a learning cycle

Until now, Vivaldi has mostly handled data collection, annotation, training and model maintenance as a service. vKI Studio now involves domain experts directly: images from ongoing inspection operations can be selected specifically by item, batch, station, time period or quality status, labelled and used for retraining.

This creates a closed lifecycle: Production inspection → real results → targeted data selection → labelling → training or retraining → comparison of model versions → renewed, approved deployment.

vKI Studio: markierte Bohrungen an einem Bauteil
Features
  • Compile image sets by item, batch, station or status
  • Mark defects and image regions in a structured way (labelling)
  • Retrain existing models with new images
  • Compare new model versions against previous ones and manage them
  • Connection to VIVALDI traceability
Inside the software

How you work with vKI Studio.

From the selected inspection image to the approved model version, all in one interface.

01 / SELECT IMAGES

Compile data in a targeted way

Images from ongoing inspection operations can be filtered by item, batch, station, time period or quality status and compiled into training or evaluation datasets.

vKI Studio: Bildergalerie mit Filtern nach Zeitraum, Material und Kameraansicht
02 / LABEL

Label defects in a structured way

Defects and relevant image regions are marked directly in the image and assigned to a defect class. This way, the expertise of your own employees flows straight into the training data.

vKI Studio: Labeling einer Pore auf einer Bauteiloberfläche
Also in dark mode
vKI Studio: Labeling im Dark Mode
03 / TRAIN

Train and retrain models

Existing models can be retrained with additional images. Training progress is visible at all times, for example via loss and F1 score.

vKI Studio: Trainingsverlauf mit Loss und F1-Score
04 / EVALUATE

Compare model versions

New model versions are checked against previous versions using metrics such as precision, recall and F1 before they are approved for deployment.

vKI Studio: Evaluation mehrerer Modellversionen mit Precision, Recall und F1
Also in dark mode
vKI Studio: Evaluation im Dark Mode
Business value

Why this matters to customers

  • No external data scientists needed for labelling and model training
  • Maximum data protection: data does not leave the company
  • Model training benefits directly from the expertise of your own employees
  • Independent optimisation of production systems without external coordination loops
  • Roles, approval status, versioning and rollback remain clearly defined
Available now

Interested in vKI Studio?

Get in touch: we will show you vKI Studio using your own use case.