Turning complex data
into economic clarity.

Policymakers needed rigorous evidence on whether economic policy uncertainty causally suppresses firm-level investment across the US economy.

Processed a 1.75 GB institutional panel of 730K+ financial records and replicated Baker, Bloom & Davis (2016) using high-dimensional fixed-effects regressions, benchmarked against Stata estimators for validation.

Traditional GDP and activity proxies lag by months in much of Europe, leaving policymakers blind to real-time regional shifts.

Engineered a convolutional neural network to classify land use across 34 European countries from satellite imagery, generating a near-real-time proxy for regional economic activity.

Do ESG-focused R&D investments actually lower the cost of capital for innovation-heavy firms – or is it greenwashing?

Built a 16-year cross-sectional panel (2010–2025) of Japanese technology firms and applied two-way fixed effects with PSM and entropy balancing to address selection bias across 818 equity and 351 debt observations.

Analysing how technology sector dominance has reshaped U.S. market structure and capital allocation over the past decade.

Processed and analysed 500K+ firm-level observations from the Compustat quarterly dataset, engineering key financial indicators to support market and economic analysis. Conducted longitudinal analysis of firm performance between 2012–2024, identifying shifts in market structure and concentration through statistical and distributional analysis.

UK Census microdata is voluminous and messy, making it difficult to quickly extract policy-relevant demographic and regional insights.

Engineered an automated Python pipeline to clean and structure 36MB+ of UK Census microdata, producing analysis-ready datasets for demographic research. Optimised data processing workflows to efficiently analyse hundreds of thousands of demographic records.

Workforce planners lacked a scalable way to see how skills overlap across industries that look unrelated on paper.

Built an NLP pipeline over 10K+ US Labor Department job descriptions to automatically classify roles and surface latent skill overlaps across sectors.

Pricing teams needed a repeatable way to monitor a fragmented product catalogue distributed across dozens of paginated pages.

Built an end-to-end scraper covering 755 SKUs across 48 paginated pages, parsing HTML with CSS selectors and automating multi-page crawling via a dynamic loop.