Projects
Building ML Systems That Deliver Real-World Impact
My project portfolio spans the full spectrum of modern machine learning and data science: from cutting-edge agentic AI platforms built with LangGraph, to production infrastructure ML systems delivering $80M+ in quantified business value, to foundational research in Bayesian decision theory and mathematical optimization.
I've transitioned from computational scientist to ML/DS professional by combining deep statistical expertise with modern frameworks (LangGraph, JAX), production engineering capabilities, and a track record of deploying systems that solve real problems at scale. Whether building autonomous AI agents, optimizing critical infrastructure with probabilistic models, or achieving 60-99% computational efficiency gains through HPC acceleration, my work emphasizes quantified impact and production readiness.
The projects below demonstrate capabilities across: agentic AI engineering (LangGraph, RAG, tool-using agents), scientific ML platforms (JAX, automatic differentiation, HPC), production Bayesian systems (uncertainty quantification, causal inference), and mathematical optimization (constrained optimization, surrogate modeling). They represent commercial deployments, research contributions, and open-source development—showcasing breadth and depth across the ML landscape.
Jump to Category:
ML Engineering
Building production AI systems with cutting-edge frameworks (LangGraph, JAX)
This category showcases my ability to build complete ML systems using modern frameworks that define the 2025 AI landscape: LangGraph for agentic AI and JAX for high-performance scientific ML. These aren't proof-of-concepts—they're production platforms deployed to solve real problems, demonstrating full-stack capabilities from architecture design to containerized deployment.
Agentic AI Document Generation Platform
End-to-end LangGraph agentic AI application with stateful reasoning loops, RAG-based document synthesis, and autonomous multi-step task execution. Built with FastAPI backend, Streamlit frontend, JWT authentication, SQLAlchemy ORM, and Docker Compose orchestration.
Impact: Production platform transforming multi-day manual document synthesis into streamlined interactive sessions | Full-stack ML engineering from UX to deployment
Technologies: LangGraph, LangChain, RAG, Agentic AI, FastAPI, Streamlit, JWT Auth, SQLAlchemy, Docker, Google Gemini
JAX-Based Scientific ML Platform
High-performance inverse problem solver using JAX automatic differentiation, constrained optimization, and flexible parameterization frameworks. Designed for HPC deployment with efficient gradient computation and scalable problem formulations.
Impact: Production-ready scientific ML library enabling complex geophysical inversions | JAX-based modern autodiff optimization
Technologies: JAX, Automatic Differentiation, Constrained Optimization, Scientific ML, HPC, Python
Quantum Gravity Sensing for Subsurface Anomaly Detection
JAX-based HPC platform processing raw quantum gravity sensor data into actionable 3D subsurface density maps. Implements multi-stage signal processing (tide correction, atmospheric correction, Gaussian Process drift modeling) and JAX-optimized inverse problem solver with Gaussian heightmap parameterization.
Impact: Production ML platform for geophysical exploration | Containerized Docker deployment for HPC clusters
Technologies: JAX, Inverse Problems, Signal Processing, Gaussian Processes, HPC, Docker, Constrained Optimization
Learn more about Quantum Gravity Platform →
Production ML
Production ML systems delivering $80M+ in quantified business value
These projects prove my ability to deploy production machine learning systems that deliver measurable, multi-million dollar business impact. From predictive maintenance optimizing $60M in infrastructure costs to automated Bayesian workflows analyzing 200+ assets, these systems demonstrate commercial value delivery at scale—not research prototypes, but deployed solutions making real operational decisions.
Queensland Rail Predictive Analytics
Comprehensive machine learning solutions for predictive maintenance of 200+ bridge assets. Implemented Bayesian networks and causal inference models to optimize maintenance scheduling, improve operational safety, and identify cost-saving opportunities.
Impact: $60M in cost savings | 200+ bridge assets analyzed | Production deployment in critical infrastructure
Technologies: Machine Learning, Bayesian Networks, Causal Inference, Predictive Maintenance, Python, Monte Carlo Simulation
Learn more about Queensland Rail project →
LXRP Asset Optimization
Led data science initiatives optimizing metro rail bridge assets, reducing costs, materials, and carbon emissions through probabilistic analysis and scenario optimization. Analyzed 1.5 million potential combinations to identify 21 critical scenarios driving design decisions.
Impact: 20% asset optimization | Reduced costs and carbon emissions | 1.5M combinations analyzed
Technologies: Data Science, Optimization, Monte Carlo Simulation, Infrastructure Analytics, Python
Learn more about LXRP Optimization →
Scalable Structural Health Monitoring Solutions
Deployed automated Python solutions for large-scale structural performance evaluation across Queensland Rail and LXRP projects. Implemented Monte Carlo simulation, Bayesian networks, and causal inference models for asset condition prediction under uncertainty with parallelized workflows.
Impact: 200+ assets analyzed | Automated reporting and maintenance optimization | Production Bayesian workflows
Technologies: Monte Carlo Simulation, Bayesian Networks, Causal Inference, Automation, Python, Parallelization
Learn more about Structural Health Monitoring →
Bayesian Methods
Advanced Bayesian methods for uncertainty quantification and decision science
Deep expertise in probabilistic machine learning, Bayesian inference, and uncertainty quantification applied to high-stakes decision problems. These projects demonstrate mastery of advanced statistical methods—Bayesian decision theory, Gaussian Processes, Polynomial Chaos Expansion—solving problems where uncertainty quantification is critical and traditional point-estimate ML approaches fail.
Model Risk & Value of Information Framework
Pioneered the first mathematical framework in India and Australia for quantifying the monetary value of data acquisition before investment, explicitly accounting for predictive model error. Developed novel Bayesian decision-theoretic approach preventing investment mis-estimations of over 200%.
Impact: $800K ROI per use case | Novel research contribution | Adopted by VicRoads for multi-million dollar transport decisions
Technologies: Bayesian Decision Theory, Value of Information, Model Risk Quantification, Uncertainty Quantification, Decision Science
Learn more about VoI Framework →
ML for Predicting Remaining Service Life (Oil & Gas)
End-to-end development of Bayesian regression models predicting corrosion initiation in steel assets for oil and gas infrastructure. Processed and merged disparate experimental datasets, performed feature engineering on numerical and categorical data, and employed sensitivity analysis (SALib) for model reduction.
Impact: 90% prediction accuracy | Significant cost savings | Uncertainty quantification for high-stakes maintenance decisions
Technologies: Bayesian Regression, Feature Engineering, Uncertainty Quantification, Sensitivity Analysis (SALib), Python
Learn more about Oil & Gas Corrosion project →
Accelerating Probabilistic Simulations with Advanced Metamodeling
Developed and validated Polynomial Chaos Expansion (PCE) metamodels reducing complex nested Monte Carlo simulation runtime by over 99% while maintaining 0.999 R² accuracy. Implemented and rigorously compared spectral methods (PCE) against Gaussian Process approaches (Kriging) for surrogate modeling.
Impact: >99% computational speedup | 0.999 R² accuracy | Transformed computationally intractable analyses into practical decision tools
Technologies: Polynomial Chaos Expansion, Gaussian Processes, Kriging, Surrogate Modeling, Computational Efficiency, Python
Learn more about Metamodeling →
Optimization
Mathematical optimization and HPC achieving 60-99% efficiency gains
Expertise in mathematical optimization—constrained, unconstrained, closed-form analytical solutions—and high-performance computing delivering dramatic computational efficiency improvements. These projects demonstrate ability to solve hard optimization problems, accelerate expensive computations through parallelization and HPC, and contribute to open-source scientific computing communities.
Parameter Optimization for Safety-Critical Systems
Replaced flawed, century-old industry heuristics with closed-form mathematical solution for infrastructure code calibration, improving design efficiency by 68% (RMSE reduction: 0.28 → 0.09). Derived analytical solution via linear algebra enabling rational calibration of billion-dollar infrastructure standards.
Impact: 68% efficiency improvement (RMSE: 0.28 → 0.09) | Closed-form analytical solution | Adoption by infrastructure standards bodies
Technologies: Constrained Optimization, Linear Algebra, Parameter Calibration, Optimization Theory, First-Principles Derivation
Learn more about Parameter Optimization →
Transport Infrastructure Performance Analytics
Created advanced regression models and probabilistic frameworks for transport infrastructure analysis achieving 99.5% computational speedup over conventional finite element approaches while maintaining prediction accuracy. Implemented surrogate modeling, structural reliability analysis, and extreme value statistics with parallelized Python workflows on HPC clusters.
Impact: 99.5% computational efficiency gain | HPC-accelerated workflows | Production deployment for infrastructure assessment
Technologies: Surrogate Modeling, Structural Reliability, Extreme Value Statistics, HPC, Parallelization, Python
Learn more about Transport Infrastructure →
PySTRA Open Source Development
Core developer of PySTRA, a Python library for structural reliability analysis. Contributed 3500+ lines of code implementing advanced statistical methods and algorithms for engineering applications, including reliability analysis, sensitivity analysis, and optimization algorithms.
Impact: 3500+ lines of code | Open-source contribution | Community support and documentation
Technologies: Open Source, Python, Structural Reliability, Scientific Computing, Algorithm Development
Explore More
Interested in discussing how these capabilities could solve challenges in your organization? Let's connect.
Want to understand my technical background and career trajectory? Read my CV or learn more about my journey.
Looking for technical deep dives? Check out my blog for articles on Bayesian methods, JAX programming, and ML engineering.