Singapore · Data analytics portfolio
Public data, checked before it becomes a claim.
Six Singapore-data analyses and one AI-assisted SQL evaluation. Each connects a question to its source, method, checks and decision memo.
These are Hermes-assisted projects. Hermes implemented much of the pipeline and presentation; Faiz approved the analytical choices and publication gates. They are not presented as unaided coding.
A town-level shift-share analysis of Singapore’s four-room resale price move. The headline median and the additive mean-based decomposition are kept separate, with window and small-town sensitivity checks.
Town-average “rate” still includes which flats sold within each town. This is not a like-for-like price index or a forecast.
A balance-versus-loss-ratio bridge on quarterly MAS data, with flow-versus-stock definitions, alternative denominators and a formula-based Excel quick-check workbook.
An arithmetic attribution is not a causal account of borrower quality. Reported cards are not unique customers.
A reference-result checker for AI-generated SQL, with a versioned question set, read-only execution and a failure catalogue. In one dated mixed-model run, 27 of 32 questions passed first try; retries rescued none of the five wrong answers.
This is a small single-table evaluation using three free-tier model IDs, not a model ranking or proof of general SQL accuracy.
SingStat indices and approximate fixed-weight contributions, with coverage and residuals disclosed. Retail and F&B stay separate; missing values are unavailable, not zero.
Exercise-level changes and definition-consistent windows. Bid counts describe auction pressure, not bidders’ values or a causal demand effect.
Paired A/B premium gaps across long and tight comparison windows. Redefined categories change the baskets; before/after differences are not a policy-effect estimate.
Same-town, same-flat-type comparisons and controlled regressions. Cross-sectional associations are not a flat’s annual depreciation rate.
Reports and code: all seven linked analyses are public. They are reviewed snapshots, not automatically refreshing dashboards. Their repositories document provenance, methods, checks and limitations.
Power BI: editable source project, screenshots and a downloadable .pbix are delivered for the HDB resale analysis. Public interactive hosting is deferred.
Tableau: the interactive retail 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.