RW Rashinda Wijethunga
Work Experience Publications Contact Download CV

About Me

I'm a Machine Learning PhD candidate at Western University (London, ON) who builds and deploys production Generative and Agentic AI systems. Recently I built BuildRight — a guardrailed, agentic hybrid-RAG commerce assistant deployed as a single Docker container — and co-built a five-agent enterprise provisioning system with tool-use, memory, and human-in-the-loop control, both backed by formal evaluation harnesses and test suites.

My PhD research centres on trustworthy anomaly detection with graph neural networks and counterfactual explanations. Through the Mitacs Accelerate program I took ML from open problem to production in supermarket refrigeration systems, cutting early slow-leak detection latency from months to days. I have 6 peer-reviewed papers (including top-tier Q1 journals), 4+ years of graduate research, and 2 industry ML deployments.

Contact: rashinda.wijethunga@gmail.com  |  LinkedIn  |  GitHub  |  Google Scholar

Research Pillars
Graph-Based Anomaly Detection

I design GNN architectures that model inter-sensor dependencies for unsupervised fault detection. My flagship work, DualSTGF, jointly learns temporal and spectral dependency graphs and uses Jensen–Shannon divergence between views as an early-warning anomaly signal — detecting faults before reconstruction error rises. DualGraphSHM fuses physical-topology and feature-correlation graphs for structural damage detection, with a 1D-CNN module that cut training time by 96%.

Counterfactual Explainability

I develop methods that make ML decisions interpretable and robust. CERTS uses Monte Carlo Tree Search to generate counterfactual explanations that remain valid even when models are retrained — addressing a critical gap in deployed systems. I've also contributed a novel conic optimization framework for plausible counterfactual generation using quasi-convex programming (CVXPY/MOSEK). My ongoing thesis work extends counterfactual explainability to spatio-temporal GNNs.

Education
PhD in Electrical & Computer Engineering

Western University, London, ON (Sep 2021 – Jun 2026)
CGPA: 92.63/100
Dissertation: Graph Neural Networks for Anomaly Detection and Explainable AI
Key Courses: Theoretical ML, Convex Optimization, Dependable AI Systems, Probabilistic Generative AI

B.Sc. Engineering

University of Moratuwa, Sri Lanka (Dec 2015 – Jan 2020)
Specialization: Electronic and Telecommunication Engineering
Honors Thesis: Indoor Navigation for the Visually Impaired

Industry & Research Experience
Graduate ML Researcher (4+ Years)

Western University (Sep 2021 – Present)
Two research streams: graph-based anomaly detection (DualSTGF, DualGraphSHM) and counterfactual explainability (CERTS, Conic CF). Led the anomaly detection component of a cross-domain collaboration on IoT-based pothole detection.

Applied ML Research Intern (2 Years)

Neelands Group via Mitacs Accelerate, Burlington, ON
Phase I (HFC, 2022–2023): Built and deployed a production leak detection pipeline on Azure — CatBoost + SHAP, CI/CD, F1 0.92. Phase II (CO₂, 2025): Designed a fully unsupervised pipeline for transcritical systems; detected 7/8 previously unknown slow leaks in validation.

Systems Engineer (1+ Years)

Millennium IT ESP, Colombo, Sri Lanka (Mar 2020 – Aug 2021)
Migrated Sri Lanka's largest telecom (SLT-Mobitel) from legacy WAN to Cisco SD-WAN and Cisco ACI data centre architecture.

Technical Skills
Deep Learning & GNNs

PyTorch, PyTorch Geometric, TensorFlow, scikit-learn, XGBoost, CatBoost, LightGBM, Hugging Face Transformers. GNN/GAT/GCN architectures, Transformers, CNN, LSTM/GRU.

Explainable & Generative AI

Counterfactual explanations, SHAP, MCTS, robustness evaluation. Convex/quasi-convex optimization (CVXPY, MOSEK). LLM-based multi-agent orchestration (Google ADK), RAG, MCP, A2A.

MLOps & Data Engineering

Azure Functions, Docker, CI/CD (GitHub Actions), Weights & Biases. Apache Spark, Polars, DuckDB. End-to-end deployment, Dash/Plotly dashboards.

Publications

6 peer-reviewed papers  |  4 published (3 Q1 journals, 1 IEEE conference)  |  2 under review/revision  |  *Equal contribution

Dual-View Spectral-Temporal Graph Networks with Adaptive Edge Learning for Incipient Fault Detection
Information Fusion (IF ~15.5) • Under review
Evaluating the Robustness of Counterfactual Generating Methods for Tree-based Ensembles
IEEE Canadian Journal of Electrical and Computer Engineering (CJECE) • Minor revision
Robust and efficient dual-graph neural networks for structural damage detection and localization
Engineering Structures, 343 (2025) • DOI
Expert evaluation system for pothole defect detection
Expert Systems with Applications, 277 (2025) • DOI
Precision leak detection in supermarket refrigeration systems integrating categorical gradient boosting
Energies, 17(3) (2024) • DOI
A Novel Objective Function for Counterfactual Explanations Using Conic Optimization
IEEE ICAC 2025 • *Equal contribution
Awards & Grants
  • Mitacs Accelerate Scholarship (Phase II) — $30,000 (May 2025 – Apr 2026)
  • Mitacs Accelerate Scholarship (Phase I) — $20,000 (May 2022 – Apr 2025)
  • Fully Funded PhD Scholarship, Western University (Sep 2021 – Sep 2026)
  • SLIOT Innovation Open Category — 1st Place (2019) & 2nd Place (2020)
  • IEEE ComSoc Student Competition — Honorary Mention, World Top 15 (2019)
Teaching
Graduate Teaching Assistant

Western University (Jan 2022 – Present)

  • ECE 9039: Machine Learning (Graduate)
  • ECE 3316: Web Technologies (Undergraduate)
  • ECE 3314: Computer Networks Applications (Undergraduate)
Community Involvement
© Rashinda Wijethunga • rashinda.wijethunga@gmail.com • LinkedIn
Built with ♥ and GitHub Pages