How Corra interviews and evaluates candidates

This statement explains, in plain language, what Corra is, how an interview works, what Corra evaluates and does not evaluate, how the integrity check works, how results are used, and what you can do if you disagree with an outcome. It is published by Correlia Inc. (“Correlia”) for candidates, for the employers and recruiting companies that use Correlia (“Employers”), and for anyone else who wants to understand the system. It is also the explanation of how the AI works and what general types of characteristics it uses that employers are required to give candidates under laws such as the Illinois Artificial Intelligence Video Interview Act, and it provides the information Employers need for notices under New York City Local Law 144, the California automated decision-making rules, Colorado’s automated decision law and the EU Artificial Intelligence Act. How your personal data is handled is described in the Candidate Privacy Notice (/legal/candidate-privacy-notice).

1. What Corra is

1.1 Corra is an artificial intelligence system developed and operated by Correlia. It conducts job interviews by video on behalf of Employers, transcribes them, and produces an evaluation for the Employer’s hiring team.

1.2 The interviewer you see and hear is not a real person. Corra’s face and voice are synthetic: they are generated by computer models built by Correlia. Nobody is watching or speaking to you live during the interview. Corra tells you this at the start of the interview, and the pre-interview screen says so.

1.3 Some Employers use Corra without the avatar. In that mode you hear Corra’s voice (or read the question) and record a video answer. Everything in this statement applies to that mode as well.

2. How an interview works

2.1 Invitation. The Employer invites you by email or through the job board or applicant tracking system it uses. The invitation tells you who the Employer is and gives you a link.

2.2 Device check and notice. Before the interview starts you check your camera and microphone, read the Candidate Privacy Notice, the Candidate Interview Terms and this statement, tick the required box, and decide whether to tick the optional box about training Correlia’s models. Nothing is recorded until you press “Start interview”.

2.3 The interview. Corra introduces itself as an AI interviewer and asks the questions the Employer configured for the role. Corra listens to your answers and may ask follow-up questions to understand your experience better, so the exact questions differ from candidate to candidate. There is no fixed time limit; the interview lasts as long as the conversation needs. For technical roles the Employer may include a live-coding exercise in which you write code in a widget while Corra discusses it with you. The whole session is recorded (video and audio).

2.4 After the interview. The recording is transcribed and evaluated, and the results are made available to the Employer’s authorised users inside the Correlia platform, usually within minutes. The Employer’s people review them and decide on next steps. You will hear from the Employer, not from Corra.

3. What Corra evaluates

3.1 Inputs. Corra’s evaluation uses: the job requirements the Employer configured (required and preferred skills, experience, knowledge and other job-related criteria); the questions and the Employer’s guidance on what good answers contain; your resume or profile as provided by you or the Employer; the transcript of your answers; and, where applicable, the code you wrote in the live-coding widget.

3.2 Outputs. For each interview Corra produces: a requirements match showing which configured requirements your resume and answers give evidence for; per-question evaluations; a hard-skills score for technical and knowledge questions; a soft-skills score based on the content of your answers; a resume score; an overall score; and a written summary that cites the parts of the transcript it relies on, so that the Employer’s reviewer can check every statement against what you actually said.

3.3 General types of characteristics used. Corra evaluates job-related characteristics that appear in the content of your answers and resume: knowledge and skills relevant to the role; relevant experience and achievements; the approach you describe to problems and decisions; the concrete examples you give; the clarity, structure and completeness of your explanations; the technical correctness and reasoning in hard-skills and coding answers; and the consistency between your resume and your answers. Scores are relative to the requirements the Employer set for the role; they are not a general judgement about you as a person.

3.4 How the evaluation is produced. Corra uses large language models and Correlia’s own models to compare the content of your answers and resume with the Employer’s requirements and the Employer’s guidance on good answers, and to write the summary with evidence citations. The Employer defines the criteria; Corra applies them consistently to every candidate for the role.

4. What Corra does not evaluate

4.1 Corra does not identify you against any face or voice database. The only face processing is the temporary same-person check described in Section 5.2; its data is deleted automatically when the interview ends.

4.2 Corra does not create a voiceprint or any profile that identifies you by your voice.

4.3 Corra does not analyse your facial expressions, eye movement, gaze, posture, gestures, tone of voice, pitch, accent, speaking speed, appearance, clothing, surroundings or background when evaluating you.

4.4 Corra does not try to detect emotions, mood, stress, personality traits, honesty, deception or health conditions.

4.5 Corra does not infer or use race, ethnicity, colour, national origin, sex, gender identity, sexual orientation, age, religion, disability, pregnancy, marital status, citizenship, genetic information, union membership, zip code or any other protected characteristic or proxy for one. Employers are contractually prohibited from configuring criteria that would do so.

4.6 Corra does not make hiring decisions, does not rank candidates against an automatic cut-off, and does not reject anyone.

5. The integrity check

5.1 Why it exists. Employers want to know that the interview was completed by the invited candidate, alone, under the conditions they set. Correlia’s integrity check gives the Employer’s reviewer a structured view of observable events during the interview instead of guesswork.

5.2 What it looks at. During the interview the platform records whether the interview window lost focus or visibility (for example switching to another tab or window, or the pointer leaving the window) and when; copy and paste events; how long you took to begin and complete each answer and how consistent those times were across questions; simple text patterns in your answers (sentence length, self-corrections, filler words); whether a person is present on camera and whether more than one person appears; camera or microphone interruptions; the language of the interview; and the approximate location (country) derived from your IP address. To confirm that the same person remains on camera, the system may also create a temporary face template during the interview; it exists only while the interview is running, is deleted automatically when the interview ends, and is not compared against any external database.

5.3 What it produces. An integrity score and a short list of the reasons behind it (for example “window lost focus during three questions” or “answer timing was unusually uniform”), with the underlying events, so that the reviewer can look at the recording at those points and judge for themselves.

5.4 What it is not. The integrity check is not a lie detector and does not measure honesty, emotion, stress or intent. It does not identify you against any external database, and the temporary face template it uses is deleted when the interview ends. It is a signal, not a finding: every signal can have an innocent explanation, such as a notification, an unstable connection, a family member walking past or a habit of pausing before answering. The integrity signals are never used to score your ability.

5.5 How Employers must use it. Correlia’s terms require Employers to treat the integrity report as one input for a human reviewer, to look at the recording before drawing conclusions, and never to reject a candidate solely on the basis of the integrity score. If a reviewer has concerns, the appropriate response is to ask you about it or to offer another conversation. If you believe a signal has an innocent explanation, tell the Employer.

6. How results are used: humans decide

6.1 Correlia is a technology provider. The Employer’s recruiters and hiring managers decide whom to advance, interview further, hire or reject. Corra’s scores, summaries and integrity report are decision-support for those people.

6.2 Correlia’s Customer Terms of Service require Employers to ensure that a person with the authority and competence to change the outcome reviews Corra’s outputs, the transcript and, where relevant, the recording before a decision is made; not to base any decision solely on an AI score or the integrity report; to offer an alternative process or an accommodation on request; and to comply with the laws on AI in hiring that apply to them.

6.3 If you are in a jurisdiction that gives you rights in relation to automated decisions (for example the EU, the UK, Quebec, Brazil, California from 2027 or Colorado from 2027), you may ask the Employer to explain the role Corra’s outputs played in its decision, to have a person review the decision, to express your point of view and to contest the decision, and to correct inaccurate data. The Candidate Privacy Notice explains how to make these requests. Correlia assists Employers in responding.

7. Accuracy, limitations and known risks

7.1 Speech recognition. Corra transcribes what you say. Background noise, a poor connection, unusual microphones, strong accents, code-switching between languages and specialised vocabulary can cause transcription errors. Reviewers can listen to the recording to check the transcript.

7.2 Conversation. Corra may occasionally mis-hear, interrupt, or ask a follow-up that does not fit. If something goes wrong technically, tell the Employer; whether an interview can be retaken is the Employer’s decision.

7.3 Evaluation. Language-model evaluations are consistent but not perfect. They can miss context, misunderstand an unusual answer, or under-value a good answer that was phrased briefly. This is why every summary cites the transcript, why reviewers are required to check the evidence, and why scores are presented as relative to the configured requirements rather than as absolute judgements.

7.4 Requirements quality. Corra evaluates against the criteria the Employer configured. If those criteria are incomplete or unrealistic, the scores reflect that. Correlia gives Employers tools and guidance to define job-related criteria and flags criteria that are not job-related.

7.5 Integrity signals. See Section 5.4. Uniform answer timing, for example, is more common among candidates who prepare carefully, and a focus change is often a notification.

7.6 Languages. Corra conducts interviews in the language the Employer selects for the role. Evaluation quality may be lower in languages for which Corra has less training and test data; Employers are told which languages are fully supported.

8. Fairness and bias testing

8.1 Design choices. Corra evaluates only the content of answers and resumes against job-related criteria; it does not use the characteristics listed in Section 4; every candidate for a role is asked the same configured questions with adaptive follow-ups; and summaries cite evidence so that reviewers can challenge them.

8.2 Testing. Correlia tests Corra’s outputs for disparate impact using the four-fifths rule and statistical comparisons of scoring rates across sex and race/ethnicity categories, on de-identified evaluation data and constructed test sets, and reviews question generation for criteria that could act as proxies for protected characteristics. Correlia will publish a summary of its bias testing, and the summary of its first independent bias audit, on this page.

8.3 Type and source of data. Corra runs on open-source large language models deployed on Correlia’s own infrastructure and on Correlia’s own models, trained on licensed and public datasets and on interview recordings for which candidates gave the optional training consent described in the Candidate Privacy Notice; no third-party AI service receives candidate data for evaluation. Speech-to-text is provided by ElevenLabs under terms that prohibit training on Correlia’s data. Evaluation uses the data described in Section 3.1 and nothing else. Correlia keeps a record of the data used to develop and test each model version.

8.4 Cooperation. Correlia supports Employers’ independent bias audits (including under New York City Local Law 144), provides the documentation Employers need for risk assessments, and updates this statement when models change materially.

9. Accommodations and alternatives

9.1 If you have a disability or another reason why an AI video interview is not appropriate for you, or if you would prefer not to be interviewed by AI, contact the Employer using the details in your invitation before starting. The Employer can arrange an accommodation (for example a different format, extra support, or an interview with a person) and Correlia’s terms require it to consider such requests. Declining an AI interview does not mean your application is closed; that is for the Employer to decide under the laws that apply to it.

9.2 Interviews have no fixed time limit, and you can pause to think before answering. Do not disclose health information to Corra; speak to the Employer.

10. Governance

10.1 Correlia maintains human oversight of Corra’s development, logs interview sessions and model versions, retains integrity and evaluation logs for the periods described in the Candidate Privacy Notice, and records every version of this statement and of the notices shown to candidates.

10.2 Correlia aligns its AI risk management with the NIST AI Risk Management Framework and is preparing Corra for the requirements that will apply to high-risk AI systems used in recruitment under the EU Artificial Intelligence Act from 2 December 2027, including technical documentation, instructions for use, logging, human oversight measures, post-market monitoring and registration. Transparency obligations under Article 50 of that Act (telling people they are interacting with AI and marking synthetic content) are implemented as described in Section 1.2.

10.3 Correlia does not use Corra, and does not allow Employers to use Corra, for emotion recognition, biometric categorisation, social scoring or any other practice prohibited by Article 5 of the EU Artificial Intelligence Act or by the laws of the places where it operates.

11. Information for Employers

Employers are responsible for the notices, consents, audits and records that the laws of their jurisdictions require of employers that use AI in hiring. To support them Correlia provides: this statement; the Candidate Privacy Notice; instructions for use and documentation describing Corra’s intended purpose, inputs, outputs, limitations and human-oversight measures; session and consent logs; the description of the type and source of data used; the retention schedule; bias-testing documentation and cooperation with independent audits; and template notices for Illinois, New York City, California, Colorado and the European Union. Employers can request these at ntitov@correlia.ai.

12. Contact and versions

Questions about this statement: ntitov@correlia.ai. Questions about a specific interview or decision: the Employer that invited you. This page records its version and effective date; previous versions are available on request.