Field-ops tools stop at a clean PDF. TerraBrains computes the engineering: 213 code-checked analyses, a true 3D ground model, and an AI that interprets it — all from the same boreholes, with the method and project inputs available for engineering review.
Bearing capacity (IS 6403 / Terzaghi-Vesic), settlement (IS 8009 / Schmertmann), piles — bored, driven (Hiley), group, lateral (Reese p-y), downdrag, well foundation (IS 2911 / IRC 45), liquefaction (Boulanger-Idriss / NCEER), seismic site class (IS 1893), automated equivalent-linear (SHAKE-type) site response with a G/Gmax–damping curve library, soil AND rock slope stability (Bishop · Spencer · Morgenstern-Price, plus Wyllie & Mah planar / wedge / toppling), seepage and FE seepage (including transient unsaturated Richards / van Genuchten flow), retaining / braced / sheet / secant / contiguous-pile / MSE / soil-nail walls, ground improvement (stone columns, vibro, deep mixing), pavements (IRC 37 · IRC 58 · AASHTO 93), tunnel convergence-confinement (Hoek), collapsible soils (ASTM D5333), and full in-situ interpretation — CPT (Robertson), flat dilatometer DMT (Marchetti / TC16) and Ménard pressuremeter (NF P94-261). The pavement family includes a time-stepping Pavement-ME / MEPDG performance predictor (month-by-month rutting, fatigue, thermal cracking, IRI and design reliability), the Burmister multilayer-elastic solver, a finite-element plate-on-Winkler slab, a two-dimensional elasto-plastic footing bearing-capacity BVP (Mohr-Coulomb + Hardening-Soil stiffness, Prandtl/Vesić cross-checked) and strength-reduction slope solvers. Code-checked to IS, Eurocode 7, NF P94-261, BS, AASHTO, ASTM and IRC. Results identify the method and available reference context, and every module is covered by the golden test suite; 203 cited suites document published-value benchmarks or formula-fidelity checks, with dedicated external-value coverage still being extended engine by engine. The whole suite has also survived a 16-agent adversarial stress-test for invalid and boundary inputs (5,199+ engine tests).
Numbers alone don't defend a report. Each module renders the plate the standard expects, and every one exports as PNG / JPG (or SVG) straight into your signed report — the figures users of legacy tools build by hand. Foundations get the 3D failure mechanism under a footing and the Boussinesq pressure bulb; slopes get the 2D critical slip circle and an orbitable 3D slip surface; seepage gets the flow net (equipotentials × flow lines); lateral piles get the p-y soil-reaction curves with deflection / moment / shear depth profiles. The lab side draws the Casagrande plasticity chart, the particle-size grading curve, the Proctor compaction curve with the zero-air-voids line, the consolidation e-log-σ′ curve, and Mohr circles with the Mohr-Coulomb failure envelope for triaxial and UCS. Differential settlement is checked against the IS 1904 Table 1 angular-distortion limits. And every analysis and test opens with a plain-English 'How it works' guide — method, inputs, governing standard, what you get.
From your borehole logs, TerraBrains fits a radial-basis-function (RBF) implicit field per geological contact — the estimator family Leapfrog and GeoModeller use — with structural-dip-aware polarity and vertical anisotropy so beds stay laterally continuous. It runs leave-one-out cross-validation (a reliability number, in metres), voxelises the volume into a block model with per-unit volumetrics (m³), and extracts smooth marching-cubes meshes you can rotate in the browser and export to CAD/BIM (GLB). It is borehole-driven, so it honours your data and tells you where it is well- or poorly-constrained; supply measured dip/orientation and it follows faulted and folded contacts too.
This is the real viewer, running in your browser — five boreholes interpolated into dipping stratigraphic surfaces. Drag to orbit, scroll to zoom, toggle units, switch to the implicit (RBF) surfaces, lift the groundwater plane, or export the mesh to GLB.
Synthetic example · generic foundation grid, no client data · drag to orbit · the same engine runs on your project data.
A reasoning model on AWS Bedrock reads your real logs plus the 3D model's cross-validation error and proposes likely unit correlations across boreholes, flags inconsistencies (outlier contacts, possible pinch-outs or offsets), and drafts a report-ready geological narrative. It is advisory by design: you accept, edit, or reject every line; it merges into your report's Soil Investigation section as an editable block; nothing is auto-signed. Every suggestion logs full provenance (model, token counts, prompt fingerprint), counts against your per-org quota, and runs behind prompt-injection defences.
A sample interpretation of the five-borehole profile above. Try the workflow — it is advisory by design: nothing is written to a report until you say so, and the engineer signs, never the model.
Grounded in the parsed logs + the 3D model’s leave-one-out error. Every run logs its model, tokens and prompt fingerprint.
The stratigraphy is laterally continuous across the site. A thin made-ground cap (0.8–1.2 m) overlies firm to stiff lean CLAY (CL), which passes into medium-dense silty SAND (SM) and then dense SAND (SP). A very dense sandy GRAVEL (GP) forms the base of the soil column above moderately weathered SANDSTONE (weathering grade III) encountered between 16.0 and 18.5 m. Bedding dips gently (~3°) toward the south-east, consistent with rockhead rising toward BH-103.
Leaving a legacy borehole-data tool? Import a compatible AGS 4.x export and review the mapping before it becomes part of your project (see /migrate). You can also import AGS 4.x, CSV/Excel, and use AI-assisted ingestion of scanned or PDF bore logs; the AI proposes the column mapping and you confirm it. Export supported project data as AGS 4, DIGGS 2.6, LandXML, IFC (BIM), GeoJSON (GIS), DXF (CAD), and a Mohr-Coulomb FE material table, with UTM↔WGS84 reprojection for survey grids.
Results carry their governing standard (IS / Eurocode 7 / BS / AASHTO) and every engine module is covered by the golden test suite. 203 cited suites document published-value benchmarks or formula-fidelity checks; dedicated external-value coverage remains an explicit engine-by-engine programme. No black-box constants.
Leave-one-out cross-validation reports the surface reconstruction error in metres, so you know which parts of the 3D ground model are well-constrained — and which need another borehole.
AI interpretation is accept / edit / reject. Nothing auto-writes to a signed report; every suggestion logs its provenance and is reviewable. Human judgement stays in the loop.
1-month free trial. Bring your AGS file or a few boreholes, build the 3D model, let the AI read it — then sign the report it feeds.