Polarok is the AI hiring intelligence that measures what people can actually do, not how a CV is worded. It scores your entire applicant pool on capability, explains every shortlist in plain language, and leaves the final call to your team. Fair by design, and ready for the EU AI Act.
The country has hundreds of thousands of open roles, yet the tools meant to fill them quietly reject the very people who could. Polarok exists to close that gap.
Applicant tracking systems were built to filter, not to understand. They match strings of text, so a qualified candidate whose CV is phrased differently, or who took a non-linear path, never reaches a human reviewer. Under volume pressure, recruiters cannot catch what the system already discarded.
In response, many teams now paste CVs into general-purpose chat models. Peer-reviewed research is unambiguous about why that fails: unconstrained language models produce inconsistent rankings and reproduce demographic bias, which is exactly what hiring law forbids. The fix is not less AI. It is AI that is structured, constrained, and accountable.
"Naive screening, feeding a CV into a general chat model, is not deployable in hiring."
University of Washington audit, 2024 . over 3 million resume comparisons
Most HR systems evaluate a person through documents and the words inside them. Polarok evaluates the person. It transforms every CV and job description into a structured capability profile, a machine-comparable view of what someone can actually do.
That single shift makes consistent, multidimensional comparison possible. Two candidates with very different CVs can finally be judged on the same evidence, against the same role requirements, every time.
Each layer does one job well, and hands clean, structured evidence to the next. Together they take you from raw documents to a defensible shortlist.
A proprietary capability ontology with language-model extraction converts unstructured CVs and job requirements into standardised, machine-comparable profiles. Skills, experience depth, task exposure, learning adaptability and role context all become directly comparable.
A hybrid engine combines semantic embeddings with a learned ranking model. Critically, it is trained only on anonymised, capability-based pilot feedback, never on historical hiring decisions, so it does not inherit the bias documented in conventional ATS and AI hiring systems.
Language-model reasoning is constrained to structured capability evidence. Protected attributes are masked and counterfactual fairness checks are applied, turning outputs into transparent explanations, ranked shortlists and interview guidance you can stand behind.
An interactive chatbot lets recruiters refine criteria in natural language, while a dynamic dashboard makes ranking logic visible. Every decision stays under human oversight, and every action is logged for audit.
No current ATS or AI screening tool combines these four layers in one deployable, SME-affordable pipeline.
This is a working preview using sample data for a Senior Product Manager role in Berlin. Click any candidate to open the explainability panel. Protected attributes are masked, and the human stays in control.
Names, photos and dates of birth are masked before processing, in line with GDPR and AGG.
Because recruitment AI is high risk under Annex III of the EU AI Act, Polarok is built compliance-by-design. This is no longer theoretical. In Mobley v. Workday, a US federal court certified a nationwide collective action over alleged algorithmic hiring discrimination. Human oversight, transparency, logging and fairness are part of our core, not bolted on later.
Name, photo, date of birth and gender are removed before evaluation, which blocks direct discrimination at the source.
The ranking model is trained only on anonymised, capability-based data, never on historical hiring decisions whose bias AI demonstrably reproduces.
Fairness tests check whether protected attributes influence the ranking. Deviations are logged and corrected during final shortlisting.
The chatbot refuses discriminatory prompts and flags proxy criteria such as "cultural fit", catching indirect discrimination.
Polarok shortlists and explains. A person always makes the hiring decision, with human sign-off required before export.
Every evaluation records the timestamp, masked attributes, fairness result and the human reviewer, ready for documentation.
Established systems are too expensive and complex for the Mittelstand. Lightweight SME tools digitise admin but offer no decision support. Polarok closes the gap in the middle.
| Capability | Enterprise ATS Workday, SAP SuccessFactors |
Polarok | SME admin tools e.g. Personio |
|---|---|---|---|
| Implementation cost | €100k to €500k+ | ≈ one tenth of enterprise | Low, admin focused |
| Capability-based ranking | Keyword led | Core function | Not offered |
| Explainable decisions | Limited | Built in | Not offered |
| Fairness controls and masking | Varies | By design | Not offered |
| EU AI Act readiness | In progress | Compliant by design | Out of scope |
| Time to deploy | Months | Rapid, SaaS | Fast |
| Full-pool evaluation | Early filtering | 100% of candidates | Not offered |
We are partnering with a small group of SMEs in North Rhine-Westphalia to validate Polarok on live recruitment workflows. Two startups have already expressed interest, and pilot feedback will directly shape the product.
Tell us a little about your team. We reply personally, usually within two business days.
Recruitment is one of the least structurally transformed enterprise domains, despite decades of digitalisation. We are building the missing decision-intelligence layer: affordable, explainable and demonstrably fair hiring intelligence for the companies that drive the German economy.
Our principle is simple. AI should automate the repetitive screening, not the human judgement. Polarok frees recruiters to focus on people and decisions, while keeping every step transparent, logged and under human control.
Bring enterprise-grade hiring intelligence to your team, with compliance built in from the first decision.