Solving the Data Bottleneck for Subsurface AI
Load existing geological models, restore them to identify faults and folding effects, then generate millions of stochastic realisations — the synthetic training data that enables deep learning for the subsurface.
From Model to Trained AI
Load & Restore
Load existing reservoir models and perform geological restoration to identify faults and folding effects.
Generate Realisations
Stochastically produce millions of geologically valid 3D model variants with full structural uncertainty.
Build Training Dataset
Export labelled geological models in industry-standard formats, ready to feed into deep learning pipelines.
Train & Benchmark AI
Train deep learning architectures on your synthetic data. Compare model performance on standardised benchmarks.
The Tools Behind the Data
Generate new geological models with GeoSketch, or load and restore existing ones with GeoEdit to identify faults and folding effects. Stochastically perturb the parameterised models to generate large training datasets. Then train and benchmark deep learning architectures with ProxySim.
What Makes This Possible
A restoration engine, a structural editing toolkit, and benchmarked deep learning — combined into one platform.
Stochastic Generation at Scale
Generate millions of geologically valid model realisations from a single template. Each realisation captures structural uncertainty in faults, folds, and surfaces.
Restoration & Structural Identification
Load reservoir models and perform geological restoration to automatically identify faults, folds, and other structural elements — enabling targeted editing and perturbation.
Model Import & Standard Formats
Load models from industry-standard formats like GRDECL. Export generated realisations back to the same formats for use in Eclipse, CMG, or any reservoir simulator.
Deep Learning Benchmarks
Multiple DL architectures trained on a large synthetic geological dataset. Benchmarked and compared so you can pick the right model for your problem.
No installs. Standard formats. Real-time feedback.
Industry-Standard Export
Output to GRDECL and other formats. Drop into Eclipse, CMG, or any reservoir simulator.
Browser-Based, Zero Install
Open a tab and start modelling. No software installs, no IT tickets.
Real-Time 3D Visualisation
See your model update live. Slice, rotate, display properties — all interactive in the browser.
Built for Multiple Workflows
Whether you need training data, uncertainty analysis, or rapid geological prototyping — the same generative engine powers all of it.
AI Training Data Generation
Create large, labelled synthetic geological datasets for training deep learning models. The data bottleneck for subsurface AI — solved.
Structural Uncertainty Quantification
Explore how changes in fault positions, fold geometries, and surface shapes affect your model. Quantify what parameters alone cannot.
History Matching
Incorporate structural uncertainties into your history matching workflow. Match history to geology, not just parameters.
Rapid Model Exploration
Load a model, restore it, and iterate on structural alternatives. Explore what-if scenarios for faults and folds in seconds.
Building the Future of Subsurface AI
GeoVoxels is developing tools to solve the data bottleneck holding back AI in reservoir simulation. Our platform combines geological modelling expertise with deep learning research to generate the synthetic training data that subsurface AI needs.
We believe that high-quality synthetic geological data is the key to unlocking accurate, fast, and scalable AI models for the energy industry. Our tools enable researchers and engineers to generate millions of geologically valid model realisations — bridging the gap between limited real-world data and the data-hungry requirements of modern deep learning.
Get in Touch
Interested in early access or have questions about our tools? We'd love to hear from you.