Singapore data-analytics portfolio: six public-data analyses and one AI-assisted SQL evaluation. The projects connect SQL and Python pipelines to data checks, charts and decision memos.
These are Hermes-assisted projects. Hermes implemented much of the pipeline and presentation; I approved the analytical choices and publication gates. They are not presented as unaided coding.
- HDB resale: rate or mix? — separates town-share mix from town-average price changes, with sensitivity checks and an editable Power BI report. Read the report.
- Card write-offs: balances or loss ratio? — an arithmetic bridge on MAS quarterly data, with an Excel quick-check workbook. Attribution is not a claim about borrower quality. Read the report.
- Can I trust AI-generated SQL? — reference-result checks, read-only execution and a failure catalogue. One dated, mixed-model run passed 27 of 32 questions first try; retries rescued none. Not a model leaderboard. Read the report.
- Retail value versus volume — fixed-weight contribution estimates with explicit coverage and residual limits. Explore the Tableau dashboard.
- COE quota and bid pressure — exercise-level descriptive associations, not causal demand estimates.
- COE category-definition break — paired A/B premium gaps across comparison windows; not a policy-effect estimate.
- HDB remaining-lease slope — controlled cross-sectional associations, not a flat's depreciation forecast.
Tools shown in the projects: DuckDB SQL, Python, matplotlib, Power BI, Tableau and Excel. Each repository documents its source snapshot, methods, checks and limitations.
Dashboard status: the Power BI source project, screenshots and downloadable .pbix are delivered; public interactive hosting is deferred. The retail Tableau dashboard is published and verified for signed-out viewing and an industry-filter test. It is a July 2026 snapshot, not an automatically refreshing report.