In this article, you'll find an explanation of what foundation models are, why DORA uses them, and how to choose the right one for your project.

A foundation model is a specialized deep learning model that analyzes geospatial data to identify patterns associated with mineral deposits. In DORA, foundation models are used to generate your Prediction Map.

Foundation models are pre-trained on globally sourced, anonymized datasets covering a wide range of geological environments and deposit types. Training involves feeding the model large batches of geospatial data repeatedly, allowing it to refine its understanding of geological patterns across many iterations. This means they arrive with established knowledge of geological and statistical relationships before your project data is introduced, allowing the model to interpret your data stack rather than simply process it.

Each deposit-specific model is trained on datasets curated for that mineral system, so its understanding of geological patterns reflects the physics of that deposit type. The Deposit Agnostic model is trained across all deposit types.

See Available Foundation Models for the full list of available models.

Generating a Prediction Map involves two separate models working in sequence.

Your chosen foundation model analyzes the relationships across your selected input layers and converts them into embeddings. Rather than evaluating each layer independently, it detects patterns and gradients across your full data stack based on what it learned during pre-training. The result is a rich, combined representation of your data that captures spatial and geological context in a format the next stage can use.

DORA then trains a separate classification model using your Learning Data. This model learns the spatial patterns associated with your mineralized and non-mineralized locations, then applies that knowledge across the AOI to generate the VRIFY Prospectivity Score (VPS) and your Prediction Map.

The foundation model supplies the spatial intelligence, and the classification model applies it to your specific project.

For a full breakdown of what happens when you run a prediction, see How DORA Generates a Prediction Map.

DORA offers three model types. The right choice depends on your project geology and how much you know about your target mineral system going in.

These models are trained on datasets curated for a particular mineral system. They develop deep familiarity with the geological signatures of that deposit type, so if you know what you're looking for and your project fits a recognized deposit type, a deposit-specific model will generally produce sharper predictions. Each model reflects the geometry and physical expression of its deposit type. For example, orogenic gold targets will tend to be linear, while porphyry targets are more distributed.

This model is trained across multiple deposit types and carries no bias toward a specific commodity or system. Because it isn't oriented toward a particular target, it can surface activity areas you may not have anticipated. It's a useful choice when you are unsure of the mineral system, want a second perspective unconstrained by deposit-type assumptions, or have limited Learning Data.

This model trains from scratch using only your project data within your selected Area of Interest (AOI). There is no prior training from other environments. Use this when your project doesn't fit a recognized deposit type, or when your geology is distinctive enough that prior training on other environments could introduce noise rather than signal.

Reach out to your dedicated DORA contact or email support@vrify.com for more information.