Privacy Notice

01 Who this notice applies to

This notice applies to the InterviewAssistAI / Interview Practice Coach Companion Windows desktop application and its related live transcription, AI practice-answer, screen-text (OCR), account and speaking-practice features. It explains what the current application implementation processes, why it is processed, where it may go, how long it is kept, and the controls available to you.

The application is a practice and productivity aid. It is not a recording-consent service, legal adviser, or guarantee that a meeting, interview, employer, platform or jurisdiction permits recording or transcription.

02 Important: only start capture when authorised

Live capture is user-initiated. It starts only after you press Start Transcription and confirm the in-app capture notice. You must use the application only where you are authorised to capture and process the audio. Before starting, tell all other people whose voices may be captured and obtain any consent or other permission required by the laws and rules that apply to them, the meeting host, the employer and the communications platform.

Your acceptance of this notice is your own acknowledgement. It is not consent on behalf of an interviewer, candidate, co-worker, customer or any other participant. It does not replace a notice or consent that must be given by those people themselves. Do not use the application for covert recording, to bypass recording indicators or platform restrictions, or to capture confidential conversations without authority.

03 Live system-audio capture

The current live-transcription backend uses Windows audio loopback to capture the system playback output of the default speaker/headphone device on the machine running the live-transcription backend. The reviewed live-transcription route does not capture a separate microphone input. It captures audio only while a user-started live session is running.

Where the backend runs matters. When the backend is running on the same Windows PC as the desktop application, the captured playback is that PC's default output. If the desktop application is configured to use a remote backend, the current capture code operates on the remote backend machine's playback device; it does not, by itself, capture the user's local Windows desktop audio. Use live system-audio capture only with a local backend on the intended device unless a separately documented local capture component has been installed and verified.

The live route receives short audio blocks, converts them to mono for speech recognition, and buffers utterances in process memory. In the reviewed implementation, these live audio buffers are passed to Faster-Whisper running in the backend process and are not intentionally written as audio recording files to disk by that live route. This statement does not cover operating-system paging, crash dumps, endpoint security tools, backups or later versions/configurations that may change the processing path.

04 Speech recognition and transcript handling

Live speech recognition is performed by Faster-Whisper on the machine running the backend. The current code loads the small speech-recognition model with CPU/int8 processing. Audio is processed to produce text; it is not sent as raw audio to the AI answer provider by the reviewed live-transcription path.

Recognised text, recent transcript context, detected questions and generated practice answers are held in the backend's in-memory live-session state while the backend process is running. Stopping capture stops new capture, but does not by itself clear the transcript and answer history. The application provides a Clear session data control after capture has stopped. A backend process restart also clears this in-memory live-session state. Clearing in-memory session state cannot recall information already sent to an AI provider, copied by the user or retained independently by a provider under its own terms.

05 AI practice answers and external providers

If AI answer generation is enabled, the application may send the recognised question and a limited amount of recent transcript context to the AI provider configured for your account or plan. The reviewed backend can route requests to a local Ollama service or to OpenAI, depending on the operator's configuration. With a local Ollama service, text is sent to the configured Ollama endpoint. With OpenAI configured, the recognised text and context are transmitted to OpenAI's API. Provider terms, retention settings, processing locations and any international-transfer arrangements must be reviewed and described by the operator before release.

You can pause/stop AI answer generation while keeping audio transcription active. When AI answer generation is paused, subsequent completed utterances are not queued for live practice-answer generation. This does not stop audio capture or transcription. Use Stop to stop the complete capture session.

The production application must identify the actual configured provider and processing location to users before use. Do not describe a configuration as local-only if it routes data to a remote server or external AI provider.

06 Screen OCR and chat

The Windows desktop app allows you to initiate a screen-region selection. The selected pixels are passed temporarily to the desktop application's OCR code, which uses Tesseract.js in the app to extract text. The reviewed flow does not automatically send the extracted text to the AI service; it places the extracted text into an editable chat composer for you to review. If you choose to send that text in a chat or AI request, the text and relevant prompt context are sent to the AI provider configured for that request.

The screen-region selector and OCR must remain user-initiated and visible. Do not select or transmit screen content unless you are authorised to do so.

07 Separate speaking-practice microphone feature

The separate Speaking Practice feature uses the browser/device microphone after the user grants microphone permission and starts a recording. The reviewed implementation uploads the completed recording to the authenticated backend for Faster-Whisper transcription. The backend writes the uploaded audio to a temporary file for processing and deletes that temporary file in its cleanup step. Temporary-file deletion is not a guarantee of forensic erasure from every operating-system, storage or backup layer. This separate feature is not the same as the live system-playback capture described in Section 3.

08 Privacy controls available in the Windows application

The desktop application provides the following controls, subject to the version and Windows APIs in use:

• Visible start and stop: live capture requires a user action; the app shows a running status and an on-screen Stop control. The Ctrl + Alt + S shortcut stops live transcription.
• Pause AI answers: AI answer generation can be stopped while transcription continues. The Ctrl + Alt + G shortcut toggles AI answer generation.
• Mask this popup on your display: this hides the popup's contents in the app UI on your own display. It does not delete the transcript or stop processing.
• Exclude the popup from supported screen captures: the app asks Windows to exclude the transcription popup window from supported screen-capture APIs. This is a best-effort Windows feature. It applies to the popup window only and is not guaranteed to work with every recorder, remote-desktop tool, camera, driver or capture method. It does not make the application invisible and does not protect other windows.
• Clear session data: after capture is stopped, you can clear the in-memory transcript, detected questions, answers and related session history in the current backend process.

These controls are not substitutes for participant notice, consent or compliance with workplace, interview, platform and recording laws.

09 Account, subscription and operational information

When you register or use a customer account, the service may process account details (such as name and email), authentication/session information, subscription and credit records, support messages and service diagnostics needed to operate the application, provide support, secure accounts and administer subscriptions. Payment processing may be performed by the payment provider presented at checkout. The operator must identify the payment provider and link to its privacy information in the production service.

The operator must document the actual hosting provider, hosting locations, AI providers, payment provider, support tools and any other processors used in production. The list may change by deployment or plan; the production privacy notice must be updated when material data recipients change.

10 Retention

• Live raw audio: the reviewed live route buffers audio in memory for transcription and does not intentionally save live audio recordings to disk.
• Live transcript and suggestions: held in backend process memory until cleared by the user or the backend process exits. Stopping capture alone does not clear the session history.
• Speaking Practice audio: uploaded to the backend, written to a temporary file for transcription, and removed by the route's cleanup step after processing.
• Account, billing, security and download-acceptance records: retained for the period needed for account administration, security, legal obligations, dispute handling and audit requirements. The operator must set and publish specific retention periods for each category before launch.
• AI provider processing: when transcript text or prompts are sent to an AI provider, that provider may process or retain the data under its applicable service terms and configured data controls. The operator must publish the provider-specific retention information before launch.

11 Sharing, hosting and international processing

Data may be processed by the operator's backend host, the configured speech-recognition runtime, the AI provider selected for answer generation, the payment provider (for payments), and infrastructure or support providers engaged by the operator. The machine running the backend may be the user's Windows PC or a remote server, depending on configuration. If data is sent to another country, the operator must assess and implement the required transfer safeguards for the relevant users and jurisdictions.

12 Your choices and requests

You can decline to download or use the application. During use, you can decline the live-capture acknowledgement, stop capture, pause AI answers, mask the popup, enable the popup's Windows capture-exclusion control, and clear the current live-session data after capture stops. You can also contact the operator using the privacy contact above to request access, correction, deletion or information about processing, subject to applicable law and any records the operator must retain.

13 Security and limitations

The operator should apply appropriate access controls, authentication, secure transport (HTTPS for remote connections), secrets management, backups, patching, monitoring and incident-response procedures. No software or network is completely secure. The Windows popup capture-exclusion feature is best-effort and must not be marketed as an absolute anti-recording guarantee. The operator should conduct a security review before distributing installers or enabling remote backends.

14 Changes to this notice

The operator may update this notice when functionality, providers, processing locations or legal requirements change. A material change should be presented to users before the affected processing occurs. The production download gate should display the current policy version and record the version accepted by the user.