PiaarFind your people

Built for trust, not for harvest.

Most AI products bolt on a privacy policy at the end. Piaar designed the matching engine around privacy in the first place. Here's what that means in practice.

Last updated: 26 October 2026 · Version 1.8
Transparency

Explainable matching

The EU AI Act asks that AI systems be transparent about how they work. That's a standard we're building toward, not one we're claiming to have already cleared — here's what's true about matching today.

  • Every match score shows the skills you and the other person have in common
  • No shadow profiling — we don't use sensitive attributes to score you
  • Email us and we'll explain how a score was calculated
  • We can tell you about a significant algorithm change if you ask
Consent-first

Consent over collection

GDPR shapes how we build, not just how we write a policy. The company is based in Sweden, our core database and server functions run in EU regions, and anything optional runs on a strict consent-based model.

  • We collect the minimum needed to match well — nothing else
  • Full user control over visibility and interactions
  • Discoverable-off in one tap; other users only ever see an approximate area, never your exact position
  • Your data leaves with you when you delete your account, aside from a short list of things that are someone else's record too — see our deletion page
Architecture

Privacy by design

Privacy-by-architecture means privacy isn't enforced by policy alone. It's something we build toward at the query layer, the storage layer, and the deployment layer — including where we still have work to do.

  • No public scraping of profiles from other platforms
  • No sensitive attributes used in scoring
  • No cross-event exposure without your explicit consent
  • Turning Discoverable off removes you from matching, lists, and the map — this protection is still being hardened across every access path, so treat it as our design intent rather than an absolute guarantee today

Where machine learning actually runs

Two places, and we'd rather name both than let you assume.

Matching. A text-embedding model (Qwen3-Embedding-0.6B, run by us on Google Cloud in europe-west1) turns each skill name into a vector, and your match score against another member is how well their offered skills cover what you're looking for. This is for networking between members only — jobs and recruiters don't use it, as the next paragraph explains. It sees skill names only — never your messages, photos, or location. Distance is a range filter and a tiebreak, not part of the score itself. A small number of promoted jobs or events can appear at the top of a list, always clearly labelled "Promoted"; nothing else affects order.

Moderation. Content you publish publicly is checked before other people see it: events, articles, event announcements, job posts, job-alert messages, and the "workplace" field on your profile. Google Cloud Natural Language reads the text; Cloud Vision SafeSearch looks at the images. Both are called through Google's EU endpoints. Your private one-to-one messages are never sent to either of these systems.

If you're Discoverable and switch on "Open to opportunities," your score against a job is worked out by fixed rules rather than by the model above. It is a count of how many of the job's skills you have: a skill you both name counts in full, a catalogue alias counts a little less, a hand-curated synonym less again, averaged across the job's skills. That same number is what a job alert's threshold is checked against, what a recruiter sees if you tap "I'm interested", and what a recruiter's own search of people ranks by — no embedding model touches any of it. Job pages show it as "X of Y skills" so you can see what it is made of. A job can list spoken languages, but they never count towards the score and never decide who gets an alert. Every job alert ends by telling you why you got it: "Sent automatically because you're open to opportunities, near the job, and have at least one of its skills." See our privacy policy for exactly what a recruiter can and can't see. A job or recruiter score helps an organization sort; it never decides who is hired. Job alerts stay limited and the recruiter features are not yet fully open.

Automated checks are wrong sometimes. When a check holds back something you published, the app sends you a notice saying why and how to appeal. Nothing is permanently deleted on a machine's say-so alone: flagged content is held for human review, and you can challenge any decision by emailing privacy@piaartech.com. A person reads every appeal, and we aim to reply within five working days.

Our commitments

Nine things we promise to do — or not do.

A short list because we want it to be checkable. If you ever find us violating one of these, write to us — we'll look into it and tell you what we find.

We will explain any match score.

Email privacy@piaartech.com and we'll send a plain-language explanation of the skills a score is based on. We aim to answer within a month, the same window GDPR sets for a data request, and sooner where we can. The rule for people is short and doesn't change: we look at how well the skills someone offers cover the skills you're looking for. A closely related skill counts for part of a match rather than nothing, because the comparison is about meaning and not exact words. Distance doesn't move that score up or down — it decides who is in range at all, and it separates people who score the same, so the nearer of two equal matches comes first. If you aren't looking for any skills yet, everyone in range scores alike and you simply see the nearest people first. Events are scored differently: mostly on what share of the event's topics are exactly among the skills you're searching for — which start as the skills you're looking for, and which you can change for a single search in Discover — with a smaller amount added for being nearby, so here distance really does raise the score. For an event the match has to be exact: a closely related skill counts for nothing, unlike on the people side. And it's a share rather than a tally, so an event with one topic you're searching for scores the whole of that part, while an event with five topics and two of them matching gets less. Two events scoring the same are ordered by whichever starts sooner, not by which is closer. If there are no skills to compare, because you aren't looking for any or you've turned on "any topic", events aren't scored at all: no score is shown, and you see every event in range with the nearest first. A job is counted differently again, by fixed rules rather than by the matching model: your score is how many of the job's skills you have, and the job page shows it as "X of Y skills". Where people and events are scored, we leave out the ones scoring too low to be worth your time — so a short list can mean a thin match rather than an empty neighbourhood. That applies the same way on Home and in Discover, in the list and on the map.

We measure the product, not you.

Usage analytics are off unless you switch them on, carry no account identifier, and are never used to rank, score, or profile you. Switching them off deletes the identifier and stops collection immediately.

We will never sell your data.

Not to advertisers, not to data brokers, and not for money to recruiters either. If you apply to a job, a recruiter does receive some of your data with your consent — see our privacy policy for exactly what. If Piaar were ever acquired, we intend that your data would transfer to the buyer only after you re-consent — that's how we'd handle it, not something our systems can enforce on their own.

We will never scrape public profiles.

Your Piaar profile is what you put in Piaar. We don't import from LinkedIn, GitHub, or anywhere else without your direct action.

We will explain algorithm changes on request.

We don't yet publish a running changelog of matching changes, but if a significant change would affect your score, email us and we'll explain what changed and why.

We will delete on request.

Deleting your account in the app removes your profile from search and matches and deletes your personal data straight away. Copies can remain in our cloud provider's recovery systems for up to 7 days, and in backups for about three months. A few things survive deletion because they're someone else's record or a legal record, listed in full on our deletion page.

We will host core data in the EU.

Our database and server functions run in EU regions. Some underlying Google services we depend on — like sign-in, push notifications and operational logs — run on Google's global infrastructure rather than an EU-pinned one. Where a processor sits outside the EU/EEA, we rely on Standard Contractual Clauses or an equivalent safeguard.

We will collect the minimum.

We ask for what a match needs — the skills you offer and seek, a rough location, and your consent. Nothing we can't justify, and nothing kept running in the background.

We will tell you about incidents that affect you.

If we experience a data incident, GDPR requires us to notify our supervisory authority within 72 hours of becoming aware of it. If it's likely to put you at risk, we'll tell you directly too, without undue delay, and explain what happened once we understand the scope.

No, thanks

Six things you'll never see on Piaar.

The features we said no to are as much a part of the product as the ones we said yes to.

×
Streaks and counters

"You've been on a 7-day streak" is a gambling mechanic with a friendlier name. We don't use them.

×
Re-engagement notifications

If we haven't heard from you in a week, we won't ping you with a manufactured reason. We trust you to come back when it's useful. The one exception: if you start signing up and don't finish, we send a single reminder email.

×
Algorithmic feeds

No infinite scroll, no engagement-optimized timeline. People are ranked by skills score, then by distance. Jobs and events can include one clearly labelled "Promoted" item — nothing else affects order.

×
Always-on tracking

Location and Bluetooth are requested only in the moment a feature needs them — never kept running in the background to follow you around.

×
Dark patterns at sign-up

No pre-checked consent boxes, no "remind me later" buttons hidden in grey. Decline is as easy as accept.

×
Cross-platform tracking

No pixel on partner sites, no shared advertising ID, no cohort analysis sold to third parties.

Reporting concerns

See something off?

If a match score looks wrong, a feature feels off-mission, or a privacy commitment looks like it's been broken — write to us. We answer everything.

Get in touch →

Privacy you can verify.

The whole product is open about how it works. Try it for yourself.

Download on the App Store Get it on Google Play