AI transparency notice & employer guide

Last updated 5 October 2026. Written with the EU AI Act (Regulation (EU) 2024/1689) in mind.

1. What the system does

PraakTark runs AI-led first-round interviews. An AI interviewer asks questions based on the role. The candidate answers by typing or speaking (speech is turned into text). After the interview, an AI model (currently Google Gemini 2.5 Flash) reads the transcript and the job description and produces scores from 0 to 10 (overall, communication, technical, answer depth), strengths, red flags, a short summary and a suggestion: "Advance to Round 2", "Consider" or "Pass".

2. Intended use

• Decision support for a trained human recruiter during first-round screening.
• Not intended to make or automatically carry out hiring, rejection, promotion or termination decisions.
• Not intended for monitoring employees, judging personality, or any use outside recruitment screening.

3. What the system does not do

• No emotion recognition. The system does not detect or infer emotions, mood, stress or honesty from voice, face, or text. No audio or video is stored. Only the text transcript is scored.
• "Confidence" or "answer depth" scores measure how specific, complete and well-supported the answers are. They do not measure emotional confidence.
• No biometric identification or categorisation.
• The scoring model is told to ignore protected characteristics such as gender, age, ethnicity, nationality, religion, disability or accent. The candidate's name, email and phone number are removed from the transcript before scoring.

4. Known limits

• AI output is probabilistic. The same transcript can get slightly different scores.
• Errors in speech-to-text, poor connections, or answers in a second language can lower scores unfairly.
• The system only sees the interview answers. It knows nothing about references, portfolio, context or how a candidate performs over time.
• Scores are not compared against a validated human-rater benchmark yet, so treat them as an indication, not a measurement.

5. Employer duties (human oversight)

• Assign a trained person to read the full transcript before shortlisting or rejecting anyone. Don't rely on the score alone.
• You may disagree with the AI. Your decision is logged, along with whether it differs from the AI's suggestion.
• Respond to candidate requests for an explanation or a human review (shown on the candidate page) within 30 days.
• Offer a reasonable alternative (such as a human interview) to candidates who ask for one, including for disability-related reasons.
• Set a data retention period on each role (30 to 730 days; 180 days by default). Data is deleted automatically after that.
• If you are an employer in the EU, inform workers' representatives where required, and keep the logs available for at least 6 months.
• Report serious incidents or suspected bias to privacy@praaktark.com straight away.

6. Candidate rights

• You are told before the interview that you are talking to an AI, and you must agree before it starts.
• From your private results page you can ask for an explanation of your result or for a human review.
• You can request a copy of your transcript and scores, or ask for them to be deleted, at privacy@praaktark.com.

7. Records we keep

For every scoring run we log the model, the prompt version, a fingerprint of the input, the output and the time. For every hiring decision we log who made it, when, and whether it differed from the AI's suggestion. These logs are kept for 12 months.

8. Contact

AI Act and compliance questions: privacy@praaktark.com.