AI in Geotechnical Engineering

Geotechnical engineering is the discipline where AI risk is most consequential. Foundations fail catastrophically when ground conditions are misread. The licensed geotech engineer’s judgment is irreplaceable. That said, AI is finding legitimate roles in soil classification, settlement prediction, and risk analysis where it accelerates the engineer’s work without replacing it.

The use cases shipping in 2026.

Soil classification from CPT data#

Cone Penetration Test data feeds classical soil-behavior-type charts (Robertson). ML models trained on regional datasets refine the classification: more accurate, more nuanced, and incorporating data sources beyond the CPT (geophysics, borehole logs, regional geology).

Production-credible. Engineer reviews; classifier accelerates.

Settlement prediction#

Empirical and theoretical settlement methods (Schmertmann, Burland, Janbu) are well-established. ML augments by:

Training on regional consolidation behavior
Incorporating non-traditional inputs (groundwater fluctuation, organic content)
Producing uncertainty bounds, not point estimates

Less mature; useful as a sanity check against classical methods.

Liquefaction risk modeling#

For seismic regions, liquefaction triggering analysis follows established procedures (Youd & Idriss, Boulanger & Idriss). ML approaches now incorporate broader data (CPT plus shear wave velocity plus regional case histories) for more refined risk.

Earns its place at portfolio scale (insurance, regional risk assessment). For project-specific work, the classical analyses still drive the engineer’s design.

Slope stability and landslide hazard#

Combining LiDAR-derived terrain, geological mapping, rainfall data, and ML produces landslide-susceptibility maps that outperform classical statistical methods. Used at regional scale (planning, hazard mapping); project-specific stability analysis still requires site-specific investigation and conventional LEM/FEM analysis.

Foundation type recommendation#

Given site conditions, structure loads, and project constraints, AI tools suggest candidate foundation types (shallow, deep, ground improvement). Engineer evaluates.

Useful as a brainstorm aid for early-concept foundation studies; not for final design decisions.

Where AI doesn’t (yet) earn its place#

Replacing the licensed engineer’s judgment. Geotech failure modes are too consequential.

Site-specific analysis without site-specific data. AI does not invent ground conditions you didn’t investigate. Don’t substitute regional models for proper investigation.

Stamped deliverables that hide the AI involvement. Transparency is non-negotiable. If AI informed a design, document it.

The integration question#

Geotech AI tools that integrate with the firm’s existing software (gINT, HoleBASE, RocScience, Plaxis, Geo5) earn their place. Standalone outputs that need re-entry are friction.

Our data engineering practice handles this integration — getting investigation data, lab results, and AI-derived classifications into the firm’s authoritative analysis stack.

What we ship for geotech firms#

For geotechnical engineering engagements:

Investigation-data ingest pipeline (CPT, SPT, lab results)
ML soil-classification with engineer-review queue
Regional-risk modeling for portfolio-scale clients (insurance, planning agencies)
Integration with gINT/HoleBASE for project-specific work
Audit trail at every AI-involved step

The professional responsibility frame#

Geotech is licensed PE work. The AI tools we deploy follow the same model as structural engineering and civil engineering generally: AI surfaces candidates, the licensed engineer makes the call, the stamp belongs to the engineer.

Geotech AI is the most cautious-adoption discipline we work in. The value is real; the discipline is essential.

Geotech AI earns its place in the analysis loop. The engineer’s judgment owns the result. Our team builds the data pipelines for geotechnical practice. Tell us about the work.