Power BI & data quality BratislavaEU hours
I build reporting systems — and the checks that catch it when something quietly stops adding up.
Available for contract work. Power BI development, data modelling, and reporting clean-up — usually where a team has more reporting than people to do it.
Remote, month to month. Fixed-price first engagement if you'd rather see the work before committing to hours. — oleksii@oleksiipodpriatov.com
What I do
Most of my work is the unglamorous half of business intelligence: getting data out of systems that were never meant to hand it over cleanly, and keeping it that way after the source changes shape.
At Swiss Re I built the reporting platform for marketing operations — around twenty Power BI reports on a dimensional model I designed, fed by a Power Query layer that pulls from SharePoint and survives when the exports change. Underneath it, a data quality framework that measures how completely the business actually fills in what it is supposed to.
Before that, two research positions at the Slovak Academy of Sciences, both attached to a thesis — real-time analysis of solar energy production, and the data platform for an energy lab running a continuous stream of IoT measurements.
The idea the work runs on
Two fields from the same completeness framework, six months apart. One climbed because the gap became visible to the people who owned it. The other fell because we stopped counting placeholders as real data.
Field completeness, March to September 2026
Share of records where the field holds a real value.
Three things worth telling
Event start date looked perfect. It turned out several teams were entering today's date as a placeholder, regardless of when the event actually ran — the field was full and the data meant nothing.
We stopped counting placeholders as filled. Completeness dropped to 83%. The metric got worse and the reporting got honest: you cannot plan around a date that just says today.
The exports feeding the model kept changing structure, and anything written against fixed column positions broke every time. I replaced that with dynamic header detection and automatic field mapping, so the pipeline reads what is actually in the file rather than what was there last month. It is the least visible thing I built and the reason nobody had to rebuild reports every quarter.
Several source fields packed multiple values into a single cell —
P&C Re; L&H Re; CorSo in one column. Standard
reporting counted that as one thing, and nobody noticed the totals were
wrong. The transformations that split and expand those into separate rows
are what made the counts trustworthy.
The data quality framework
The piece I am most attached to isn't a dashboard. It measures how
complete the underlying business information really is — per field, per
manager, per business unit — and it treats placeholders like
N/A and - Choose - as missing rather than filled,
which is where most completeness metrics quietly lie.
That last number is the whole argument. The field wasn't mandated or chased. It became visible next to someone's name, and it filled itself.
Tools
Some of my own code is public — a reconciliation tool that checks every paid order actually reached the warehouse, and names the ones that didn't.
Where I've been