Joburi Studenți
Our own job community for students and young graduates in Romania — a website and groups with more than 50,000 active members — now on a platform we rebuilt, with AI that helps people write and never judges them.
Visit joburistudenti.com- Client
- Internal product
- Year
- 2026
- Status
- Ours and running — the rebuilt platform is moving to joburistudenti.com
- 50,000+
- Active members across the site and its groups
- 86,000
- Spam pages on the old site, retired with a 410
- 2
- Languages, with translated URLs in both
What it is
Joburi Studenți — “student jobs” — is where students and young graduates in Romania find their first work: part-time jobs, summer jobs, internships, first jobs after university. It is ours. We own and run the website and its groups, which together have more than 50,000 active members.
Employers post there because that is where the students are, and students come back because the jobs are real and new ones arrive every day. Most of the work is keeping the community alive: the groups useful, the listings honest and the scams out — so students find work, and companies find the best new colleagues.
Why we rebuilt it
The community had outgrown the site it ran on. The old WordPress site had stopped working in every way that matters, so we rebuilt the platform ourselves — and used it to try ideas about AI we would want to see on any platform that handles people’s personal data.
What was broken
Every public page returned a server error. Signup had been failing for everyone. And through an unthrottled signup form — whose captcha answer was sitting in a hidden field on the same page — it had collected around 86,000 spam accounts, each with a public profile page, all indexed by Google under the brand.
A patch was not going to fix that. It needed rebuilding, without losing what the brand had earned.
What we built
A new platform, Romanian first and English second, with translated URLs in both. Students sign up, build a profile, search and filter, apply with a CV, track their applications and get an email when an employer decides. Employers create a company, post and edit listings, read applicants and mark them reviewed, accepted or rejected.
Pages that show data are rendered on request and held in the CDN until something on them changes — then exactly those pages are dropped. A listing is live the moment it is published and gone the moment it is taken down, with no redeploys. Every job carries the structured data Google Jobs reads.
The old site’s URLs were mapped across with redirects. The 86,000 spam pages were not: they answer 410 Gone, the strongest signal there is to drop them, because passing their reputation on to real pages would be backwards. Company pages now have to be earned by an approved listing.
AI that helps people write, and never judges them
This is the part we are proudest of, and most of it is about what the AI is not allowed to do.
For employers:
- Paste a post, get a draft. A Facebook post or an email becomes a filled-in job form in about a second. The model only picks out which lines are the listing; the description is then cut from those lines word for word, so the employer’s words stay theirs. Every field it returns is checked, and a pasted “ignore previous instructions” was ignored in testing. Nothing is published by an import — it returns a draft to read.
- Review my listing. Before publishing, it lists what to fix: a requirement about age, sex or appearance (quoted, with the sentence rewritten without it), text that contradicts the form, pay mentioned in the text but missing from the salary fields, a title that does not say what the job is. The model reports; the code decides — a quote that is not really in the listing is thrown away.
- Translate between Romanian and English, faithfully, with numbers and names untouched.
For students:
- Feedback on an application message: what works, what to improve, and what the listing asks for that the message leaves out. It is the student’s tool — shown only to them, never stored, never seen by the employer, no score. It judges the message, never the person. Emails, phone numbers and links are stripped out before anything is sent, and it never writes facts for the student: in testing, models kept putting the listing’s requirements in the student’s mouth (“I know Canva”), so the design changed — an example sentence is only shown if every word in it is the student’s own.
For everyone:
- A scam check. Students are a favourite target of fake job ads — a “training fee”, crypto, a recruiter who only talks on WhatsApp. When a listing is about to go public without an admin looking, the model looks for red flags from a fixed list, with a quote as evidence. Clear flags hold the listing for review; it holds, it never rejects, and the employer only sees “waiting for approval”.
The lines we drew
- Listing text and candidate text never meet the same model. Anything a student writes goes only to an AI provider hosted in the EU, and if none is configured, the feature simply is not there.
- No AI ranks people. Matching candidates to jobs with a model is what the EU AI Act treats as high-risk. “Jobs for you” is plain rules — your city, remote if you said so, newest first — with the reason shown.
- CVs are deleted on a schedule. Each application keeps its own copy of the CV, readable only by that student and that employer, and it is deleted 180 days later. Deleting an account deletes its applications and files at once.
Where it stands
The community keeps growing in its groups. The new platform has been running since 4 October 2026, with the old blog posts migrated and every flow working end to end; moving it onto joburistudenti.com is next.
- Community
- Job board
- Applied AI
- Platform rebuild
- GDPR
Knowledge this took
- Running and moderating a large online community
- Full platform rebuild — Astro, Postgres, accounts and roles
- Applied AI with a hard line on what data may leave
- Designing AI features around the EU AI Act
- Prompt-injection-resistant extraction and moderation
- CDN caching with exact invalidation
- SEO recovery — redirects, 410s, structured data for Google Jobs
- GDPR-led data retention for CVs
- Romanian-first localisation, English second
So we can also build
- Marketplaces and job boards where trust is the product
- Turning a community that lives in social groups into a platform it owns
- Adding AI to a platform without putting users' personal data at risk
- Rescuing a legacy WordPress site that has been taken over by spam
- Multilingual platforms where one language is the contract
Something close to this but not quite it?
Got something like this in mind?
Ask Ola how we'd approach it, or go straight to a person.