A Sort-of Case Study from Inside Hungary’s Largest Job Board

unsplash-image

As a freelancer I spent 7 years at Profession.hu, Hungary’s largest job board. I touched pretty much everything: B2C search experience, B2B job posting flows, the job ad editor, CV tools, internal moderation systems, customer support tools; all of it, overlapping, messy, interconnected.

I’ve been meaning to turn my experience into a portfolio piece for ages, but it’s not a clean story or a straight line. This is more about how I worked as a designer, what problems we tried to solve, and why we approached them the way we did.

The theme that kept coming back, again and again: recruiting platforms live or die on data quality. Everything else is downstream from that. But getting good data means forcing users to do work they absolutely don’t want to do.

Employers hate filling forms. Candidates hate filling forms. Moderators hate dealing with forms no one filled.

That tension never goes away. All you can do is try to design around it. Here's how approached a couple of core areas:

 ***

Progressive Filtering

Job search was broken. Type “marketing”, get 200+ results, and be on your own. We had filters, but people didn’t use them. Or they overused them and ended up with zero (0) jobs.

The default industry approach: show all filters upfront and hope users willingly do the labor. But the more I looked at our own data, explicit (search queries) + implicit (behavior), the more obvious it became:

We already know enough to guide people. Why are we making them do all the work? So I pushed for progressive filtering:

  • You search “marketing”: 200 results. Too much. Don't show results yet.
  • Suggest experience level, location, qualification or whatever we knew that would meaningfully narrow the search.
  • Do this until we hit the "goldilocks zone". For jobs with specific requirments this was maybe 5-10, for others it might be 30-40.

For the first time we were opinionated. Historically we just surfaced whatever we had: 0 jobs, 400 jobs, good luck. Now we said: “Give us these two answers and based on years of data, this is the most useful next step.” And you could always ignore it and explore manually. 

This wasn’t about being clever; it was about reducing cognitive load while respecting user intent.

***

Optimizing for Opt-Out

One of the design principles I kept pushing was: Help candidates realize a job isn’t for them before they apply.

It sounds ridiculous: fewer applications? But if you think about it:

  • Candidates hate being rejected.
  • Recruiters drown in noise.
  • Companies don’t have infinite time to read bad applications.

For some roles (blue-collar, high-volume), sure, you want quantity. But for most? You want fit, not volume.

The real pushback came from sales, especially for big clients. Everyone loves the idea of “more applications” because it looks good on paper. But that only works if humans magically gained more hours.

Helping people self-disqualify is weirdly one of the kindest things you can design.

 ***

CV Parsing

The CV database is a core revenue stream; employers pay to search structured candidate profiles. But candidates overwhelmingly uploaded PDFs and left their profiles empty. The structured part remained a wasteland.

We tried everything: shorter forms, smarter forms, chunked form. None of it mattered.
Users already did that work when they made their CV.

We tried AI parsing a couple of years ago but it was basically useless.
But when LLMs got good, it finally made sense to try again.

We optimized the prompts to extract only the stable stuff: names, dates, company, role. Pre-filled what we were confident about. Asked for review where judgment mattered.

It wasn't perfect but it moved the needle.

 *** 

Job Listing Quality

Job listings were often low-quality: missing salary, perks, location, clarity. All the things candidates care about.

Obvious solution on paper: force recruiters to fill everything out.

Real-world problem: recruiters are buried, not lazy. They want to find candidates, not fight forms. Add to that:

  • every employer has a different idea of what a job description “should” look like
  • big paying customers often had the worst-quality listings
  • revenue vs. quality is a real tension, not a theoretical one

Block them with required fields? You lose deals. Let them post garbage? Candidates suffer.

So instead, I pushed for post-publication quality improvement:

  • moderation tools
  • human moderators (now AI-assisted)
  • structured nudges instead of hard blocks
  • feedback loops instead of rejections

Again, same pattern: design for the reality of users’ constraints, not for the idealized behavior you wish they had.

 *** 

Design is working with constraints

I couldn’t prove ROI on a lot of these decisions. Profession.hu wasn’t great at metrics, and I didn’t have the political weight to demand dashboards or control KPIs.

But I could embed myself in the problem. And that turned out to be the advantage.

I saw how candidate experience connected to employer tools, connected to internal processes, connected to moderation and customer support workflows. Or more honestly, how they didn’t connect. I learned that, “the obvious solution” almost never works because it ignores one of:

  • the mental model of the users
  • the constraints of the organization
  • the economics of the business
  • the internal politics
  • the messy multi-stakeholder dynamics

As a contractor with no direct authority, influence came from understanding all of that and designing within those constraints, not pretending they didn’t exist.

Recruiting is messy. There are no clean wins. There’s just moving the needle: reducing friction, clarifying intent, getting out of users’ way, nudging the system toward better quality without blocking people.

Say hello