AI · DATA SCIENCE · STRATEGY CONSULTING
I sit at the intersection of
technical depth and strategic clarity.
TECHNICAL DEPTH
I build models. AI, causal, predictive.
STRATEGY CONSULTING
I advise executives on AI adoption and strategy.
END-TO-END DELIVERY
I take projects from ambiguity to delivery.
TECHNICAL DEPTH
I build models. AI, causal, predictive.
STRATEGY CONSULTING
I advise executives on AI adoption and strategy.
END-TO-END DELIVERY
I take projects from ambiguity to delivery.
ABOUT
Most people who understand a RAG pipeline can't explain its strategic implications to a CEO. I can do both.
Six years across strategy consulting and applied data science, leading teams and managing clients including Microsoft, Google, and Uber — building the technical work and presenting the findings, end to end.
We're at an unusual moment — AI is moving faster than most organisations can make sense of it. The gap between what the technology can do and what decision-makers understand is wide, and widening. I've never fit neatly into either the technical or the strategic box, which turns out to be useful right now.
- CURRENTLY
- Data Science Manager, Mandala Partners
- PREVIOUSLY
- Accenture Strategy / AlphaBeta
- EDUCATION
- Australian National University · Bachelor of Advanced Computing (Honours)
- BASED
- Sydney, Australia
EXPERIENCE
Mandala Partners
Data Science Manager
June 2024 – Present
Coalition for an Insurable Future: led a county-level panel regression in Python quantifying climate risk's impact on US home insurance premiums, with state/year fixed effects, hazard–reinsurance interaction terms, and forecasts to 2035 and 2050 under multiple IPCC scenarios. Findings covered by Reuters, Politico, and regional US media.
Jobs & Skills Australia: built a two-stage skill classification pipeline (vector embeddings + LLM reranking) mapping LinkedIn-derived skills into the Lightcast taxonomy, evaluated against a human-labelled gold set across multiple LLMs. Separately built a RAG pipeline over ASX 200 annual reports to extract structured AI investment data. Outputs informed national workforce policy.
Uber Eats Taiwan: designed the analytics for a delivery work value study — OLS regression on 15,000+ delivery partners quantifying drivers of hourly earnings, paired with a randomised survey experiment eliciting willingness-to-pay for flexibility (75% preferred flexible hours over a higher-paying fixed schedule).
Accenture Strategy
Strategy Manager and Data Scientist
Sep 2019 – May 2024
AlphaBeta Australia was acquired by Accenture in 2021
Built a natural experiment causal model for a major US delivery marketplace using 8 historic platform outages as exogenous supply shocks. Compared observational, experimental, and quasi-experimental identification strategies; decomposed effects through downstream lateness, ETAs, and 12-month customer GMV.
Built a worker lifetime value (LTV) model in Python for a major US delivery marketplace; used by the client to identify which worker segments were most valuable and the drivers of that value. Counterfactual framing estimated per-order worker impact across fulfilment quality and zone-hour supply impact.
Built a difference-in-differences model for NSW Treasury evaluating three COVID-recovery voucher programs (Dine & Discover, Stay NSW, Parents NSW); continuous-treatment DiD on LGA-level panel data with LGA and month fixed effects. Findings published in NSW Treasury's program evaluation report.
Analysed a Discrete Choice Experiment for NBN Co modelling willingness-to-pay across plan attributes; estimated A$15.7B in annual consumer surplus from the national broadband network. Released as a public report.
WORK
THE WORK
County-level panel regression in Python with state and year fixed effects and hazard-reinsurance interaction terms, forecasting climate risk's impact on US home insurance premiums to 2035 and 2050 under multiple IPCC scenarios. Integrated FEMA, US Treasury, Census, and reinsurance index data across 3,000+ counties.
THE IMPACT
Findings covered by Reuters, Politico, and regional US media; framed the public debate on insurance market exposure to climate risk.
THE WORK
Two-stage skill classification pipeline mapping LinkedIn-derived skills into the Lightcast taxonomy: vector embeddings for candidate retrieval, LLM reranking for final selection. Evaluated against a human-labelled gold set across multiple LLMs and prompt architectures. Separately built a RAG pipeline over ASX 200 annual reports to extract structured AI investment data.
THE IMPACT
Outputs informed national workforce policy. Two-stage approach beat single-prompt LLM classification on the gold set.
THE WORK
Natural experiment quantifying the financial value of supply availability at zone-hour level. Used 8 historic platform outage events as exogenous supply shocks. Compared observational (distributed lag), experimental (holdout), and quasi-experimental identification strategies before settling on the natural experiment as most robust.
THE IMPACT
Decomposed effects through downstream impact on lateness, ETAs, and 12-month customer GMV. Outputs informed the client's time-to-accept targets and cost-per-incremental-delivery thresholds.
THE WORK
Discrete Choice Experiment modelling consumer willingness-to-pay across broadband plan attributes (speed, data, inclusions, price). Separately built a multivariate regression model isolating 'naked' broadband prices to benchmark Australian affordability against 12 OECD peers.
THE IMPACT
Estimated A$15.7B in annual Australian consumer surplus from the national broadband network. Both studies released as public reports.
SKILLS
BUILD
I build models — predictive, econometric, and AI.
ANALYSE
I extract insight from complex, messy data.
COMMUNICATE
I make technical findings land with executive audiences.