AI Practices

1. You are interacting with AI

Prep chat, Live Tutor, interview practice, group discussion, Office agents, job-fit write-ups, résumé and cover-letter drafts, Skill Lab explanations, and voice sessions are produced by artificial systems. They are not human recruiters, faculty, or lawyers. When a character has a name or a voice, that is a design, not a person.

Images, slides, board drawings, and spoken audio generated in the product are machine-generated. Do not present them as a human-created work or as an official exam or offer letter.

2. What we send, and to whom

To run a feature you start, we send the minimum context that feature needs. That often includes profile excerpts, résumé text, the job description you selected, your recent messages, and, for voice, the audio stream.

• OpenRouter, Inc.: language-model inference for coaching, evaluation, and generation.
• Deepgram, Inc.: speech-to-text and text-to-speech when you use a voice session. Default voice provider.
• Tavily: web search queries from prep and research tools.
• ElevenLabs: only if that voice provider is enabled on the server.

Do not paste secrets, other people's data you cannot share, or health or government-ID numbers into chat or voice. Provider processing follows their terms. We do not sell your prompts. We do not train a Career Matrix foundation model on your content.

3. Human control and intended use

• Outputs are suggestions. You decide what to send to an employer or college.
• You can interrupt the Live Tutor, stop a voice session, edit generated documents, and forget coach memory.
• Job-fit scores, mock-interview marks, and readiness meters are practice signals, not a hiring or admission decision.
• Institutions must not use ICM outputs as the sole basis for placement, admission, grading, or discipline.
• We do not use emotion recognition to infer your feelings in a workplace or educational decision.
• We do not operate social scoring, untargeted face scraping, or real-time remote biometric identification.

4. India

Digital Personal Data Protection Act, 2023 (DPDP)

AI processing is optional processing based on your consent (the AI checkbox at sign-up or invite accept). We describe purpose, processors, and withdrawal in the Privacy Policy. Children under 18 are not permitted. We do not behaviourally advertise to children.

IT Act and IT Rules, 2021 (as amended 2026)

ICM is a computer resource. We publish this notice so users can tell synthetic output from human output. We prohibit unlawful content in the Terms. The Grievance Officer is support@career-matrix.com: we acknowledge grievances within 24 hours and aim to resolve them within 7 days, including complaints about synthetically generated information that impersonates a real person.

Responsible AI practice we follow

• Purpose limitation: models run for the career-development feature you invoked.
• Human review before you rely on an output outside ICM.
• No deceptive presentation of AI as a human examiner or recruiter.
• Abuse and rate limits on generation and voice endpoints.
• A reachable grievance channel in India (support inbox above).

5. European Union, EEA, and United Kingdom

EU AI Act (Regulation (EU) 2024/1689)

Conversational and generative features are limited-risk systems with Article 50 transparency duties (enforceable 2 August 2026). This page, in-product labelling, and the sign-up AI consent satisfy those duties: you are told you are interacting with AI, and generated media is machine-generated.

We are a deployer of third-party general-purpose models (via OpenRouter), not the provider of those GPAI models. GPAI documentation and training-data summaries are the model provider's obligation.

ICM is built for you to practise and prepare. It is not placed on the market as a high-risk employment, admission, or exam-scoring system under Annex III. We do not use it to decide access to education, to filter job applicants for an employer, or to allocate workplace tasks. If a college tried to do that with our scores, that use is forbidden by our Terms.

GDPR

Lawful basis for sending content to model and voice providers is consent, plus contract where the feature is the service you requested. You may withdraw consent. We do not make solely automated decisions with legal or similarly significant effects.

6. United States

• We align internally with the NIST AI Risk Management Framework (Govern, Map, Measure, Manage) and the OECD AI Principles (updated 2024): transparency, human-centred values, robustness, and accountability.
• We do not make consequential automated decisions about employment, credit, housing, or insurance. California ADMT pre-use notices for significant decisions therefore do not apply to ICM scores.
• We do not sell or share personal information for advertising. AI processors receive data to run the feature, which is a service-provider / processor disclosure, not a sale.
• Voice audio is not used to create a biometric identifier. We do not enroll Illinois (BIPA) or similar voiceprints.
• FTC Act: we do not claim that AI output is guaranteed to get you hired, or that a synthetic interviewer is a human.

7. Quality, safety, and incidents

• Generation and voice routes are rate-limited. Uploads have size caps.
• LLM errors returned to the browser are generic. Provider payloads are not echoed.
• Generated SVG/HTML is sanitized before display.
• If a serious incident involving an AI feature harms a person or breaks the law, email support@career-matrix.com. We will investigate and, where a regulator must be told, we will tell them.

8. Changes

If we add a processor that receives personal data, or we change the intended use in a way that could become high-risk under the EU AI Act, we will update this notice and ask for new consent where the law requires it.