AI Processing Disclosure

Only these features send your content to an AI provider. Every one of them is clearly marked in the interface before you run it:

Every other tool on the site is either pure browser JavaScript or ordinary server-side processing with no AI involved.

We use Google Gemini through the Google Gen AI API. The specific model is configurable by our administrators and defaults to a current Gemini Flash model; the exact model used for each request is recorded with that request.

Content you submit to an AI feature is transmitted to Google and processed under Google's Gemini API terms. We do not use your content to train any model, and we do not send it to any AI provider other than Google.

PDF extraction runs as a staged pipeline that escalates only as far as it has to. Pages with a real text layer are extracted directly, with no AI and no image ever leaving the server. Scanned pages go to a local OCR engine (Tesseract) that also runs on our own server. AI is used only to repair text those stages could not resolve.

If a document can only be handled by sending full page images to the AI provider, we do not do it silently: you are quoted for it and must give explicit consent before that path runs.

• Text-to-speech runs entirely inside your browser. Both the basic voice and the high-quality local voice generate audio on your device; no audio and no text is sent to us or to any AI provider.
• Image upscaling in the AI Image Studio is a conventional Lanczos resample performed on our server with the Sharp image library. It is not an AI super-resolution model and we do not claim it reconstructs detail. Only background removal in that tool uses AI.
• The deterministic parts of the résumé checker — contact details, section detection, formatting and keyword checks — are ordinary code. AI is used for the written advice on top of them.

For each AI call we record the feature, the user it belongs to, the model name, input and output token counts, the computed cost, and the timestamp. We do this to track spend and price features honestly.

We do not store the prompt, your uploaded content, or the generated output in those records.

• AI output can be confidently wrong. Transcripts mishear words, translations drift, OCR invents plausible characters, and summaries omit things that mattered to you.
• Always check AI output against your source before acting on it, sending it to someone, or publishing it.
• Never treat it as legal, medical, financial, or professional advice.
• The résumé compatibility score is our own estimate of how machine-readable a résumé is. It does not replicate or predict any specific employer's applicant-tracking system.
• Arabic, mixed Arabic-English content, dialects, handwriting, and poor scans are all harder cases where quality drops noticeably.

Please do not upload material you are not allowed to share with a third-party processor — government-classified documents, protected health records, full payment card data, or other people's personal data you have no right to disclose. If you are unsure whether a document may leave your organisation, use the browser-based tools instead, which never upload anything.

Uploaded documents are treated as untrusted data. If a file contains text that tries to instruct the model — a hidden line telling it to ignore its task, for example — that text is content to be processed, not a command to be obeyed. We do our best to enforce that, but no defence is perfect; another reason to read the output before trusting it.

If you need more detail about how a specific feature handles your data, ask us at [email protected].