Independent data work
Client projects and my own practice on real exports. It started with working through my own data and moved to client work in 2026.
- Cleaning a popular-science portal's user base of bot registrations: labels derived from snapshot diffs, a leakage found in the original framing, threshold chosen by cost of error, and stability checked over time. Error analysis exposed a defect in the labels themselves — the list went to the owner and the base was cleaned further against it.
- Catalogue and order analytics for an online store: a mart of 2,971 products and 4,234 orders, funnel analysis by status, assortment and price analytics.
- A method for computing delivery metrics on cleaned data: explicit measurement boundaries and percentiles instead of a mean, against an inflated median in the raw export.
- A data-quality monitoring service: hourly checks of 24 external sources, field contracts, tests, CI, and a dashboard.