Artificial Intelligence
Interactive AI systems for modelling, inference, and visualization.
Hi, I'm Javad Ghorbani
I build interactive, explainable systems for geotechnical intelligence, autonomous constitutive modelling, and scientific visualization—bridging physics, data, and design.
- Computational Geomechanics (FEM, HM/THM)
- AI for Engineering (RL, MCTS, Bayesian)
- Explainable Visualization (D3.js, Three.js)
- Data & Knowledge Graphs (xAI)
Selected Projects
Each interactive animation below highlights a key project. Click to explore the full experience.
Computational Geomechanics
Physics-based and numerical geomechanics. Project list coming soon.
About
Bio
Applied scientist and computational geomechanics engineer building production‑grade, explainable AI for the ground. I combine physics‑based modelling (FEM, HM/THM multiphysics, unsaturated soils, contact mechanics) with modern ML (reinforcement learning, surrogate modelling, Bayesian inference, MCTS/PSO) to automate constitutive model formulation, parameter identification, and real‑time ground assessment. I design human‑centered visual analytics that turn complex simulations and data streams into actionable engineering decisions.
- Computational geomechanics: FEM, large‑deformation elasto‑plasticity, SANISAND
- AI/ML for engineering: RL, Bayesian inference, surrogate modelling, MCTS/PSO
- Inverse modelling & calibration: parameter identification, uncertainty quantification
- Data & knowledge: PostgreSQL, knowledge graphs, xAI, provenance
- Interactive visualization & UX: D3.js/Three.js dashboards for engineers
Skills
- Python
- C++
- TypeScript
- React
- D3.js
- Three.js
- PyTorch
- TensorFlow
- NumPy/SciPy
- Pandas
- SymPy
- FEM
- HM/THM
- Constitutive Modelling
- Algorithmic Differentiation
- Reinforcement Learning
- Bayesian Inference
- Surrogate Modelling
- MCTS
- PSO
- PostgreSQL
- Knowledge Graphs
- Neo4j
- xAI
- Docker
- CI/CD
- Chart.js
Contact
Interested in collaborating or learning more? Reach out and let's chat.
Links
Multiagent system for automotact consitituve model formulation, implementation and calibration (GenCAI)
Step-by-step visualisation of the orchestrated agents for automatic constitutive model formulation, implementation, and calibration.
Research Pillars
Four interconnected capabilities drive Spider's intelligent geotechnical problem-solving.
Symbolic AI meets Machine Learning - Where physics-based rules and data-driven models collaborate for reliable geotechnical decisions.
Typed Knowledge Graph (Σ)
Explainable Provider Selection
Hybrid Rules ↔ ML Pipelines
Physics-anchored numerical methods - Advanced finite element workflows with constitutive models for real-world soil and rock behavior.
Constitutive Law Implementation
Uncertainty Quantification
Multi-scale Model Calibration
Unified data foundation - From historical archives to live sensor streams, ensuring data quality and temporal consistency across all sources.
Real-time Sensor Integration
Automated Quality Control
Canonical Symbol Mapping (Σ)
Parameter recovery from observations - Using Bayesian inference and optimization to identify governing parameters from field measurements.
Bayesian Parameter Estimation
Closed-loop Model Updating
Uncertainty Quantification
System Deep‑Dive
Watch the SPIDER engine transform any input into confident, auditable decisions
What you're about to see is the SPIDER engine at work. The animation visualises how Spider ingests anything, translates it into a shared geotechnical symbol set (Σ), chooses the best providers, and explains the result.
- Typed Σ symbol set
- Provider registry
- Assurance traces
- Physics‑first checks
Intelligent Geotechnical Solver
Spider's intelligent geotechnical solver that dynamically selects optimal computation paths, combining physics-based equations, empirical correlations, and ML models to solve complex soil mechanics problems with full traceability and explainability.
Integrating climate, traffic, soil, and monitoring data for road networks
Australia’s road network faces growing distress from extreme rainfall, heat, heavy freight. A siloed view misses interactions that drive failures like rutting, cracking, and pavement moisture damage. We are working on an integrated approach fuses climate, traffic, soil mechanics, AI and monitoring data to provide a shared context for each segment, enabling earlier detection, targeted maintenance, and more resilient design.
Weather
Traffic
Model
Climate & Traffic
Daily rainfall (mm)
Temperature (°C)
Max wind gust (km/h)
Traffic Flow Index
Ground Assessment with Reinforcement Learning
Digital co‑pilot for QA/QC: fuse midpoint measurements with physics‑aware RL to estimate void‑ratio distributions and uncertainty, iteratively learning from observations.
📊
Measurements
Soil parameters, environmental data, and site conditions
Measurement
Collect soil measurements at grid midpoints
→
📊
Estimation
Predict void ratio for measured cell
→
🧠
Learning
Update policy from measurement vs prediction
📈
Void Ratio Analysis
Optimized void ratio distributions
Measurement Grid (3×9)
Void Ratio Analysis
Measurement
Estimation
Learning
AlphaGeo — Constitutive Modeling Automation and Automatic Calibration
AlphaGeo is a multi‑agent system that automates traditionally demanding tasks in geotechnical modeling—constitutive model setup, parameter identification, and calibration. In this demo, AlphaGeo: (1) samples parameters and initial conditions within bounds (Explorer), (2) builds a surrogate dataset from states S and deltas D (Self‑Play), (3) trains a model f: X→ΔA to predict parameter adjustments (Learner), and (4) refines adjustments with MCTS to minimize calibration error.
Explorer Agent (adjust A, G within bounds)
Explorer samples A (model parameters [a₁..a_M]) and G (initial conditions [g₁..g_NIC]) within their bounds. Each sample yields a prediction yᵢ and a delta Δyᵢ = yᵢ − y₀ (vs baseline y₀). States are S = {(x, y₀)} and deltas are D = {Δyᵢ}; together they form X = {S, D} for the learner.
Self‑Play Trials (build X={S,D}, Y=ΔA)
X collects states S and differences D where S = {(x, y₀)} are baseline input‑output pairs, D = {Δyᵢ} are per‑trial prediction deltas; A are model parameters [a₁..a_M]; G are initial conditions [g₁..g_NIC].
Learner Agent (train surrogate f: X → ΔA)
MCTS (selection → expansion → simulation → backprop)
Final Parameter Finding
Repositories
Everything is developed in the open. Explore the codebases below and get involved via issues and discussions.
Related publications
Knowledge-driven Geotechnical Reasoning with Symbolic Inference
Describes the Spider approach to unifying domain rules, physics, and machine learning in a reproducible pipeline.
Trustworthy AI for Infrastructure Engineering
Outlines assurance cases, provenance tracking, and standards alignment for safety-critical decisions.
Research team
Engineers and researchers building explainable geotechnical intelligence.
Dr. Alex Example
Principal Investigator — Knowledge Systems
Alex leads Spider's knowledge-driven stack: typed ontologies, assurance cases, and hybrid reasoning pipelines. Previously at ExampleLab and Uni Example.
- PhD, Geotechnical AI — Uni Example
- Focus: symbolic inference, provenance, safety cases
Jamie Taylor
Lead Engineer — Symbolic Reasoning
Jamie designs the rule engine and provider registry for auditability and speed, focusing on typed KGs and verifiable explanations.
Casey Morgan
Research Fellow — Geotechnical Modelling
Casey bridges field data and computational mechanics, calibrating constitutive models and building uncertainty-aware checks.