Upul Senanayake — Data & AI Executive

§ 01

Three careers, one thread

The Scientist

Rigour, first

A PhD in Artificial Intelligence at UNSW, using deep learning on longitudinal brain MRI to catch the earliest signs of dementia, with publications in IEEE ISBI, PLoS ONE and Chaos. Twelve years lecturing at the University of Sydney taught me that if you can't explain a model to a room of sceptics, you don't understand it yet.

The Builder

Systems that ship

At Westpac and Macquarie I moved ML out of notebooks and into production: foreclosure models that cut manual reviews by 90%, a self-service platform 150+ practitioners use daily, feature stores and gRPC services holding p99 under 200 ms. Platforms outlive projects; I build platforms.

The Executive

Outcomes, owned

Now I sit where technology meets the deal: US$12M in enterprise fraud-analytics engagements closed at Mozn, budgets to A$7.5M managed, teams grown from 6 to 12, standards adopted across every client deployment. Sharpened by Harvard Business School executive education in business strategy and design thinking, the craft is converting technical depth into commercial conviction.

§ 02

Experience

3 markets · 15 clients US$12M closed 5M+ txns/mo <20ms p99

Own the fraud analytics strategy and delivery roadmap across behavioural biometrics, device intelligence, adaptive authentication and fraud decisioning for enterprise financial-services clients across 3 markets and 15 clients.
Closed 2 enterprise deals worth US$12m by leading post-qualification technical discovery and solution architecture, translating platform capabilities into commercial propositions.
Took user-profile similarity and behavioural-entropy models from design to production, now scoring 5m+ transactions per month in real time for authentication and transaction-risk decisions.
Delivered streaming ML capability spanning mobile telemetry, Kafka ingestion, real-time feature engineering, feature stores and inference services, with p99 inference latency below 20 ms.
Established standards for model validation, explainability, drift detection and controlled releases, adopted across all client deployments and the fraud engine.
Grew the fraud analytics capability from 6 to 12 people, aligning product, engineering and fraud specialists around reusable delivery frameworks and operational risk controls.

4 operating groups 90%+ adoption 150+ enabled weeks → <1 day

Led data engineering and applied machine learning across the horizontal technology platform supporting Macquarie's four operating groups.
Defined and enforced enterprise data architecture and engineering standards across dimensional and Iceberg lakehouse modelling, lineage SLAs and automated data-quality testing; adopted by more than 90% of new data products.
Built a self-service analytics and ML layer on AWS, Dataiku and Presto/Iceberg, enabling 150+ analysts and data scientists and reducing data-to-model lead time from weeks to under one day.
Delivered LLM-enabled service-desk automation using GPT/BERT and approximately 100k tickets per year, increasing first-touch resolution by 22% and eliminating five FTE of manual triage.
Productionised incident-risk scoring, capacity forecasting and CMDB anomaly detection through feature-store and gRPC services with p99 latency below 200 ms.
Built and led a 12-person cross-functional squad across data, ML and analytics engineering; established review and on-call practices, maintained attrition below 5%, and mentored 40+ practitioners through communities of practice.
Authored a model lifecycle playbook covering feature engineering, validation and drift/bias monitoring, improving average model AUC by approximately 0.04; chaired architecture and engineering working groups with 50+ stakeholders.

A$7.5M budget ~A$20M savings case

Developed the enterprise data science strategy and target operating model, covering data platforms, ETL, analytical workbenches, MLOps and delivery governance.
Managed an A$7.5m annual budget and developed a predictive demand-planning business case with potential annual savings of approximately A$20m.

−80% handover −60% time to prod

Built and led a team of four senior data scientists and ML engineers, partnering with the founder and C-suite to establish operational ML capability for a late-stage neurotechnology company working in connectomics (structural and functional brain mapping).
Defined and implemented the MLOps roadmap and model CI/CD capability, aligned to Series C funding dynamics, reducing data-science-to-engineering handover overhead by 80% and time to production by 60%.

15 AI/ML initiatives A$10M+ forecast savings −90% manual reviews

Led a portfolio of 15 AI/ML initiatives and partnered with C-suite and GM stakeholders to establish teams, governance and production pathways for machine learning.
Drove the shift from project-centric delivery to reusable data products, addressing technology and change-management constraints and forecasting annual savings above A$10m.
Developed a predictive foreclosure model that supported earlier customer intervention and reduced manual reviews by 90%.
Devised individual price-elasticity models to set optimal interest rates for term-deposit renewals, alongside ML for acquisition, attrition, credit risk and multimodal customer analytics.
Shaped customer-centric data pipelines and the infrastructure blueprint for operational machine learning across the bank.

Open banking / CDR

Designed an open-banking personal-finance platform using Consumer Data Right data to aggregate accounts, enrich transactions and generate actionable customer insights.
Developed commercial banking and insurance use cases based on machine-learning models trained on transaction-level data.

§ 03

Expertise

Leadership

Data & AI strategy · operating models
Portfolio governance · budgets to A$7.5M
Executive & C-suite engagement
Product leadership · commercial architecture
Team development · squads of 12+, 40+ mentored

Platforms & Delivery

AWS · Dataiku · Presto · Iceberg
Python · SQL · PySpark · R
Kafka · streaming & event ingestion
Feature stores · real-time inference · gRPC
MLOps · CI/CD · Docker · Kubernetes

Applied AI

Fraud analytics · behavioural biometrics
Predictive ML · risk · forecasting · elasticity
Generative AI & LLMs · GPT/BERT in production
Responsible AI · explainability · drift & bias
Model governance · lifecycle playbooks

§ 04

Research & Teaching

Executive Education

Harvard Business School

2024 · CERTIFICATES

Certificate in Business Strategy and Certificate in Design Thinking & Innovation: case-based HBS programs on strategic trade-offs, value creation and innovative problem-solving.

PhD, Artificial Intelligence

University of New South Wales

2014 — 2019 · SYDNEY

Deep learning for longitudinal multimodal brain MRI: a deep fusion pipeline for early diagnosis of mild cognitive impairment, and survival analysis of progression to dementia.

MPhil, Complex Systems

University of Sydney

2013 — 2014 · SYDNEY

Developed the p-index, a PageRank-based alternative to the h-index for fairer measurement of scientific impact.

BSc, Computer Engineering

University of Peradeniya

2008 — 2012 · KANDY, SRI LANKA

First Class Honours. ML-driven search-space optimisation for drug discovery, adopted into the AutoDock Vina pipeline.

Senior Lecturer

University of Sydney

2014 — 2025 · 12 YEARS

Business Statistics, Business Analytics, AI in Business, Project Planning, Scheduling and Analytics.