My job discovery pipeline (remote-job-engine) sweeps dozens of ATS boards and aggregators every day. Until recently its output was a CSV I opened in a text editor, which is fine when you are hunting for one job and useless when you want to see the shape of the whole market.
So on 17 November I froze one sweep: 15,516 live postings from 73 sources, covering 1,638 distinct companies and board slugs. Then I built the same dashboard twice, once in Tableau Public and once in Power BI, because that is the fastest honest way to learn where the two tools actually differ instead of guessing from feature lists.
The dataset
One flat table. Columns: title, company, location, remote, posted, source, url. No joins, no star schema, which is the point. A BI demo is not impressive because the modelling is exotic, it is impressive because the numbers survive scrutiny.
The pipeline tags remote when the location or title signals remote work. On this snapshot that is 2,086 of 15,516 rows. Titles containing junior, graduate or entry account for 362. Those are the figures I would defend: they come straight from the CSV that ships in the repo, not from a screenshot I describe.
Building the same thing twice
In Tableau Public the four sheets are roles by source, top companies, roles over time, and a large remote-share number. In Power BI the same four visuals sit behind measures written in DAX: total roles, remote roles, remote share, distinct companies, distinct sources, plus junior roles and a "roles with date" measure.
Two things stayed with me.
First, the measure definitions are the product. A bar chart of roles by source is three lines in either tool. Deciding that remote share is SUM(remote) / COUNT(*) and then checking that the denominator counts rows rather than non-blank cells is where the correctness lives. I committed both definitions (the Tableau calculated fields and measures.dax) so a reviewer can inspect them rather than take my word.
Second, the tool differences show up at the edges rather than the core. Both swallowed a 15,000-row CSV in seconds. Tableau's calculated-field syntax feels closer to a spreadsheet; DAX forced me to be explicit about row context and filter context. Neither was harder. They are differently opinionated, and I would choose based on what the team already runs.
What the data actually said
The uncomfortable part is worth more than the dashboard. Only 5 of the 15,516 postings matched both remote and an explicitly junior, graduate or entry title. I had been running a job-hunting pipeline for months, and the snapshot made the scarcity visible in one glance: the market is not full of openly junior remote roles, it is full of senior ones. That reframed how I read every posting afterwards.
A dashboard that makes you update a belief is doing its job. One that only confirms what you already thought is decoration.
Known limits
Two, both honest. The posted column came back empty in this snapshot, so the time-trend sheet has nothing to plot until the pipeline starts capturing dates. And the junior signal is keyword-based, so it misses phrasing like "early careers" or "first role". I left both visible rather than hiding them, because a report that admits its own gaps is easier to trust than one that looks finished.
Where this sits
The pipeline finds my shortlist. This dashboard turns one day of its output into evidence that I can model data, define measures, and ship a report in two mainstream BI tools. For a junior data analyst application that is the difference between telling someone you use Tableau and handing them a workbook they can open.
Top comments (1)
"The measure definitions are the product" is the line I'd put in front of every new analyst. Checking whether the remote-share denominator counts rows or non-blank cells is exactly the kind of thing that quietly changes a number by a few points and nobody notices until a stakeholder does. I also like that you committed measures.dax and the calculated fields to the repo, so the logic can be reviewed like code. Did you hit any spot where Tableau and Power BI gave slightly different numbers for the same measure? That's usually where I learn the most about a tool's defaults.