Eleven Job Titles, One Job

Delivery Lead. Program Manager. Head of Operations. Projektleiter. Same work, eleven different job titles.

A script showed me that in early August. It went through 84 job boards and pulled 8,520 postings. If you search for one title, you miss the rest: 102 of roughly 370 relevant postings.

That script turned into BewerbungsGenie. As of today it is open, as a public beta, for Austria.

Why I built it at all

It started with irritation about CVs. Writing one that fits you is work already. Tailoring it to every single posting is the same work again, every time.

For what comes after that, I found nothing that was any good. Not for interview practice, and not for AI help with applications and cover letters. A chat assistant does not keep your profile, does not know Austrian conventions and has no reason to stick to evidence.

Then there was a practical requirement. It had to work on a phone as well as on a laptop, with AI as the foundation, not a button bolted on afterwards.

The August script then showed me that writing is only half of it. Finding the jobs already goes wrong.

What works today (and what is still coming)

BewerbungsGenie runs in the browser and is split into areas that build on each other. The interface is in German, documents come out in German or English. Four areas work today:

• Profil (profile). You upload your CV as a PDF and it becomes your profile. Everything else reads from that one profile, so you never type anything twice.
• Finden (find). Jobs in Austria, matched by role rather than by job title. Every match lists its must-have requirements. For remote and foreign postings, a traffic light shows whether they are open to Austria, unclear, or not.
• Bewerben (apply). CV and cover letter for the posting, as a PDF. German cover letters use the formal Sie.
• Vorbereiten (prepare). The interview coach asks about the job and your profile, you answer out loud and afterwards you see what already works. Next to it is the photo studio: a phone snapshot becomes an application photo. You pick from up to three variants, each labelled as AI-generated.

Still coming is Verwalten (manage), an overview of your open applications. It does not exist yet, and the app says so instead of showing an empty table.

The one hard rule: no PDF without proof

Advice on AI-written applications warns about one thing above all: getting caught. A language model will write experience you never had into a cover letter without hesitating.

So BewerbungsGenie has one rule nothing gets past. Every sentence in your CV and cover letter has to trace back to an entry in your profile. The proof check (Belegprüfung in the app) tests that before a PDF is created. Whatever it cannot back up gets flagged, sentence by sentence. You add the evidence to your profile or you drop the sentence. As long as anything is open, the PDF stays locked.

The check runs in three steps. First a fixed lookup whether the source exists in the profile at all. Then a model that checks numbers, periods and titles against the profile text. Last, a guard without any AI. It lets words like "seasoned" or "well-versed" through only with evidence (the model had waved them through as mere rephrasing).

The part that matters is easy to overlook. Sometimes the model returns no verdict for a sentence at all. A missing verdict is not an acquittal, though. In code it is a single comparison:

const verdict = verdicts.get(unit.id);
if (verdict?.verdict !== 'belegt') {
  return [{ text: unit.text, note: verdict?.note?.trim() || undefined }];
}

Anything that is not explicitly belegt (proven) fails, including a verdict that never arrived. Otherwise a silent model would be enough to open the lock. And the first instruction in the PDF export is assertFactCheckPassed(). No passed check, no file, whatever the screen happens to show.

The same attitude shows up in two other places. Upload a scanned CV and you get no text recognition, just a plain message that the PDF has no text layer. OCR would be quick to add, but it produces exactly the kind of profile that looks plausible and is wrong. I have written about a related kind of silent invention before. A transcription model once wrote an endless loop of politeness into a file for me.

And for the knock-out criteria of a posting, say a required driving licence, the verbatim sentence from the ad sits right next to it. The quote is checked against the ad text. No quote, no criterion.

For me, verification is the actual work anyway. Here it just lives inside the product.

Where it is still weak

Postings come, as they did for the script, from employers' own applicant tracking systems. That makes Finden strong today for tech, digital and scale-up roles and weaker for classic office, retail and care jobs. I am not putting a date on when that changes.

Beta also means I keep building, and things can still break. Interview coach and photo studio can be switched off individually. If the coach is paused, the app says so instead of promising a date. And Verwalten is missing, see above.

Pricing and trying it

Free costs nothing. It includes matches with the traffic light and must-haves, one CV and cover letters per posting. The PDF has no watermark, and you get two practice interviews a month and three application photos. On Free, all AI steps share 40 requests a month.

Premium is 12.90 euros a month, cancel any month. It adds what Free cannot do: tailoring your CV to each posting, with a list of the changes. Karriere-Boost is 34.90 euros once and gives you everything in Premium for three months, no subscription.

You can try it at bewerbungsgenie.at. If something looks off, write to hallo@bewerbungsgenie.at. I read those mails myself.

How my prototypes turn into products at all, I wrote up here. The other things I build are in the product overview.