CV Analyzer — Privacy Notice & Acceptable Use

Effective: 2026-07-28. Last reviewed: 2026-07-28. Owned by HUMANMAX AI PTE. LTD. ("BetterPay.ai").

1. Plain-language summary

• You upload a resume (PDF or DOCX). We extract the text in memory and analyse it with AI.
• We do not store your original file or its full text.
• Before sending text to the AI provider, we automatically redact name, email address, phone number, postal address, and personal profile URLs.
• We do not use your CV, or anything derived from it, to train or fine-tune AI models — ours or anyone else's. See section 7.
• We retain only structured findings — skills, role, seniority, education, employment history (titles & companies), gap analysis. These are kept until you delete them — there is no automatic expiry.
• You remain in control: view, correct, or delete any analysis at any time from your dashboard. Deletion is immediate.
• This feature is available to free and Pro users; Pro receives gap analysis, suggestions, and market fit. Daily AI token quotas apply.

2. What we collect when you use the CV Analyzer

• The uploaded file, in memory only, for the duration of the analysis request. The file is never written to disk, never sent to S3 or any blob store, and never logged.
• A SHA-256 hash of the file (an opaque identifier that lets you recognise re-uploads — it cannot be reversed back into the file).
• The redacted text, transmitted to our AI provider (OpenAI) for the sole purpose of producing a structured analysis.
• The structured analysis output (skills, technologies, employment history with titles and company names, education, certifications, languages, gap analysis, suggestions). This is what we store.
• A numeric embedding generated from the anonymised summary of your analysis. This is a vector of numbers used to rank job matches; it is derived from already-redacted content and is not readable text.
• Job match records, if you opt in to job matching (see section 9): the job matched, a fit score, the reasons, and an AI-generated insight.
• Operational metadata: file MIME type, file size in bytes, timestamps, model identifier, token count, redaction counters.

3. What we DO NOT store

• The original uploaded file (PDF/DOCX bytes).
• The extracted full plain text of your resume.
• Your name, email address, phone number, or postal address.
• LinkedIn / GitHub / other social profile URLs that were present in the document.

4. Why company and institution names are kept

Company names and educational institutions are employment and qualification facts, not direct personal identifiers, and they materially improve the accuracy of seniority assessment, market comparisons, and career path advice. If you do not want them retained, do not use this feature.

5. Our role, and who is responsible

Generative AI features involve several parties. In the terms used by Singapore's Personal Data Protection Commission (PDPC), BetterPay.ai is a System Deployer: we do not build or train the underlying model. We procure it as a service (via API) from a third-party Model Provider and we build the CV Analyzer around it ourselves.

That means we bear primary responsibility for ensuring this feature meets our obligations under the PDPA, including the personal data our system newly creates or collects — your uploads, the analysis outputs, the embeddings, and the match records. You should come to us first with any question or complaint; we will not redirect you to our AI provider.

Before adopting an AI provider we assess the safeguards it documents upstream — data access controls, data residency, retention behaviour, encryption, and whether inputs are used for training — so that we can judge whether the end-to-end system is adequate. We re-assess when we change provider or model, and we will not adopt a provider that cannot give us enough information to make that assessment.

Because we are a deployer and not a model developer, the obligations that attach to building generative AI models do not apply to us. We do not web-scrape personal data to train models, we do not assemble training corpora, and we do not rely on the PDPA's "publicly available" exception to do either. What applies to us are the deployment-stage duties — purpose limitation, protection, accountability, retention — and the post-deployment duty to answer your access and correction requests.

6. AI sub-processing

We send the redacted text to OpenAI for analysis. OpenAI processes the text per its enterprise API terms (no training on inputs by default). The response is parsed into JSON on our server, stored as the analysis record, and returned to you.

We may switch AI providers in the future. We will update this page if we do, and your previously stored structured findings will not change format.

7. We do not train AI models on your data

Your CV, the text extracted from it, the structured findings, your embeddings, and your job match records are not used to train or fine-tune any AI model — not ours, and not our provider's. We use AI models only to run the analysis and matching you asked for.

If we ever want to use your data for model training or fine-tuning, we will not rely on this notice or on a general "product improvement" clause elsewhere in our terms. We will tell you specifically what data would be used, for what model and function, and ask you to opt in first. You will be able to decline without losing access to the CV Analyzer.

8. What we use your data for — and what we don't

We process the personal data in your CV only for the purposes below, and we collect only what those purposes need:

• Producing your CV analysis (skills, seniority, gap analysis, suggestions, market fit).
• Matching you against job listings, if you have opted in to job matching.
• Enforcing plan entitlements, rate limits, and daily AI token quotas.

We do not:

• Sell your data, or share your analysis with employers, recruiters, or other users.
• Use it for advertising, credit, insurance, financial profiling, or background screening.
• Feed it into unrelated features, or query it for purposes beyond those listed above.

We also collect as little as the purpose allows. Data minimisation is built into the feature rather than left as a policy promise: the original file is discarded once text is extracted, direct identifiers are stripped before the text reaches the AI provider, and we persist structured findings instead of your resume. Each of those choices removes personal data we would otherwise be holding. Where we have judged that retaining something is necessary — company and institution names — we say so openly in section 4 rather than quietly keeping it.

9. Automated job matching

Job matching is opt-in and off by default. If you enable it, a scheduled job periodically compares your stored analysis against new listings for the positions you are watching, scores the fit, and generates a short AI insight for the strongest matches. It runs without you being present.

• It only reads your own analysis and your watched positions. It has no access to other users' data, and it cannot reach systems outside its defined purpose.
• Its output is a recommendation for you to read. It never applies to a job on your behalf, never contacts an employer, and never makes a decision about you.
• You can turn matching off at any time, and dismiss or delete individual matches.
• Deleting a CV analysis also deletes every match record derived from it.

10. How we protect this data — and who is accountable

A generative AI feature creates categories of personal data that a normal web form does not: the text you submit, the model's output about you, and the derived data in between. We treat all of it as personal data belonging to you, and we have assigned it an owner rather than leaving it to sit between teams.

Accountable owner. BetterPay.ai's engineering lead is responsible for safeguarding CV uploads, analysis outputs, embeddings, and job match records, and for reviewing the controls below. Questions and complaints reach that owner via security@humanmax.ai.

The specific safeguards in place today:

• The raw file never lands anywhere. It is held in memory for the request, never written to disk or object storage, and never written to logs.
• Redaction happens before any external call, so direct identifiers do not leave our server in the first place.
• Uploads are verified by content, not by claim. We check the file's actual type from its magic bytes and reject files whose declared type contradicts their contents.
• Every read and write is scoped to the owning account. Analyses and matches are queried by user, so one account cannot reach another's data even if an identifier is guessed.
• Deletion cascades. Removing an analysis also removes its embedding and every match record derived from it, in the same operation.
• Sessions are cryptographically signed, and the scheduled matching job authenticates with a secret compared in constant time.
• Per-user rate limits and daily token quotas cap how much can be extracted through the feature in bulk.
• Data is encrypted in transit and at rest.

Staff access. A small number of authorised administrators can see operational metadata — how many analyses an account has, when they were uploaded, and the detected current role — for support and abuse monitoring. They do not browse your skills, employment history, or gap analysis. Administrative access is permission-gated, rate-limited, origin-checked, and written to an audit log.

We review these safeguards as described in section 15, and we will publish material changes here rather than change them silently.

11. Retention & deletion

• Analyses are retained until you delete them. There is no automatic time-based expiry.
• Why we keep them indefinitely rather than expiring them: the analysis is the input to ongoing job matching. New listings appear continuously, so an analysis that expired on a timer would silently stop producing matches and force you to re-upload your CV — which means handling your raw file again. Keeping the structured findings means we process the sensitive original less often, not more.
• We periodically review stored analyses to confirm retention is still justified, and we may delete analyses belonging to accounts that have been closed or dormant for an extended period.
• You can delete any analysis manually at any time. Deletion from our live database is immediate, and cascades to the embedding and all derived match records.
• Any residual copies in database backups are purged as part of our standard backup rotation.

12. Your rights (PDPA)

• Access — you can view your analyses in the app, in the form in which we hold them.
• Correction — if the AI got something wrong (for example, the wrong seniority, a misattributed employer, or a skill you do not have), you can correct it. The fastest route is to delete the analysis and re-upload, which regenerates the findings and replaces the derived embedding and match records. If that does not resolve it, email us and we will correct the record directly. We handle correction requests case by case, and we will tell you if we decide a correction should not be made and why.
• Erasure — delete any analysis from the UI, or contact security@humanmax.ai for full deletion.
• Portability — request a JSON export of your structured analyses.
• Withdrawal — stop using the feature; existing analyses can be deleted on request.
• You may complain to the Singapore PDPC if you believe we mishandle your data.

How long we take. We aim to acknowledge access and correction requests within 5 working days and to resolve them within 30 days. If a request will take longer, we will tell you why and give you a date.

Because we do not train models on your data, correcting or deleting an analysis fully removes it from everything downstream — there is no trained model left holding a copy. This is the main reason we can offer correction as a real remedy rather than a best-effort one: the findings live in an ordinary database record, not baked into model weights. In the narrow cases where the PDPA permits us to decline a request (for example, where the burden of complying would be unreasonable or disproportionate to your interests), we will tell you which exception we are relying on and why.

We also try to prevent inaccuracy rather than only correct it after the fact. The analysis is derived from the document you supply, so it reflects a source you control; we hash each upload so repeat uploads of the same file are recognisable rather than silently duplicated; and we record which model produced each analysis and when, so a finding you dispute can be traced back to how it was generated.

13. Acceptable use

Only upload what the analysis needs. A CV for this tool needs your work history, skills, and education. It does not need, and you should remove before uploading:

• NRIC, FIN, passport, or other government identification numbers.
• Bank account, payment card, or salary slip details.
• Health, medical, disability, religious, or family information.
• Photographs of yourself, date of birth, or marital status.
• Referee names and contact details, and any other person's contact information.

Our redaction is a safety net for identifiers that slip through, not a licence to upload sensitive data. Anything you leave in is text we have to process.

You also agree NOT to:

• Upload a CV that is not yours, unless you have explicit consent from the data subject.
• Upload documents containing other people's personal data, classified information, intellectual property you do not own, or content prohibited by law.
• Use the analysis to make automated decisions affecting third parties (e.g., automated hiring rejection) without independent human review.
• Use this feature to profile, screen, or vet another person.
• Attempt to recover redacted PII from the analysis output, reverse-engineer our redaction, or probe the system for security weaknesses.
• Submit content designed to manipulate the AI model (prompt injection) targeting other users.

Violation may result in suspension of access and, where applicable, reporting to authorities.

14. Limitations & accuracy

• The analysis is generated by an AI model and may contain errors, including confident-sounding statements that are simply wrong. Treat it as guidance, not professional advice — and see section 12 if you want something corrected.
• Redaction is best-effort. We use regex + heuristic detection. We strongly recommend you also remove your address, ID numbers, and other sensitive details from your CV before upload.
• Maximum file size: 4 MB. Supported formats: PDF, DOCX. Daily AI token quotas apply.

15. How we keep this current

AI capabilities and the risks that come with them change quickly. We review this notice and the safeguards behind it at least annually, and whenever we change AI provider or model, add a materially new use of your data, or extend the system's ability to act on its own. Material changes will be notified in-product before they take effect.

This notice is written to align with Singapore's PDPA and takes into account the PDPC's Advisory Guidelines on Use of Personal Data in AI Recommendation and Decision Systems, together with the PDPC's Proposed Advisory Guidelines on Use of Personal Data in Generative AI (issued 2 June 2026, public consultation closed 1 July 2026). The proposed guidelines are advisory, still in draft, and may change before they are finalised; we will revisit this page when a final version is issued.

16. Contact

Privacy questions: security@humanmax.ai. General support: support@humanmax.ai.