Kanvis

Privacy Policy

Effective Date: May 17, 2026

At Kanvis ("Kanvis," "we," "us," or "our"), we believe your professional story deserves the best tools to tell it. This Privacy Policy explains what we collect, how we use, store, share, and protect your information when you use our AI-powered candidate portfolio platform and related services (the "Service"). By using the Service, you consent to the practices described in this policy. Words used here in initial caps without definition have the meanings given in our Terms of Use.

1. Information We Collect

A. Information You Provide

  • Account Information: Name, email address, profile picture, and authentication identifiers when you sign in via Google, email verification code, or another supported method through Firebase Authentication.
  • Professional Documents: Resumes (PDF/DOCX), cover letters, profile media, and any other files you upload. Uploaded files are stored in Google Cloud Storage and may be parsed by document-parsing providers (e.g., LlamaParse) and large language model providers (e.g., Google Gemini) to extract structured profile data.
  • Profile Data — the "Talent Graph": Everything that builds up your portfolio, including:
    • Identity (full name, location, summary, tagline, username once claimed, optional profile picture);
    • Narrative (professional summary, motivations, core values, working preferences);
    • Experiences, education, skills, projects, awards, certifications;
    • Recommendations, social profiles, profile media (highlight links/images), working styles, FAQs;
    • Executive-level entities where applicable (governance roles, philanthropy, intellectual property, external activities).
  • Conversational Data: Full transcripts and metadata from your interactions with our AI agents — including Azmuth (the candidate-side Build agent) and the AI Twin (the public-profile read-only chat). Each chat message is stored in an immutable audit ledger (ChatMessage) so we can link every change in your Talent Graph back to the conversation that produced it.
  • JD Mentor Inputs: Job descriptions you upload (PDF/DOCX) or paste, the parsed JD, a frozen snapshot of your profile at the time of the analysis (profile_snapshot), and the resulting score breakdown and long-form analysis.
  • Generated Artifacts: STAR-format resumes (PDF and LaTeX bundles) generated by the Service, recruiter outreach drafts (ShareTemplateHistory), and any other AI-produced output saved to your account.
  • Onboarding Progress: Items in your in-product onboarding checklist (OnboardingChecklist) used to guide profile completion.
  • Authentication Auxiliary Data: Email verification codes (issued and consumed during sign-in) and recruiter-waitlist sign-ups.
  • Published Profile Content: Everything visible at kanvis.me/<username> once you publish.

B. Information Collected Automatically

  • Session & Device Data: Browser type, operating system, device identifiers, IP address, and Firebase session identifiers (including the temporary anonymous UIDs we create before you sign in).
  • Usage Analytics: Pages visited, features used, time spent, click patterns, interaction sequences, and engagement metrics — captured through PostHog and our own server-side logging.
  • AI Interaction Metadata: Input/output sequences, tool calls executed on your behalf, conversation flow, interaction directives, gamification deltas, and completion rates.
  • Profile View Analytics: Every visit to your published profile creates a ProfileView record (page, timestamp, coarse referrer information) used to power your dashboard's view counts, "views this week," all-time views, and percentile metrics.
  • Referral Attribution: When a visitor lands via a referral link (?ref=<your-username>), we attribute the referral to power your dashboard's influence/peer-count metrics.
  • Performance Data: Error logs, exception stack traces (captured through PostHog), response times, AI service latency, and diagnostic information.

C. Anonymous Session Data

You can begin using the Service without creating an account. When you do:

  • We create a temporary, anonymous Firebase Authentication session and associate your profile-in-progress, uploaded documents, chat history, JD analyses, and other data with that anonymous UID.
  • This data is processed on our servers and may be retained on our infrastructure for the purposes described in this policy.
  • When you later sign in with an email/Google account, the anonymous session is linked into your authenticated account so your progress carries over.

D. Cookies and Local Storage

We use cookies and browser local storage to maintain your session, remember preferences, and capture product analytics. We provide an in-product cookie-settings modal (accessible via the footer) where you can review and adjust your consent for non-essential categories. Essential cookies (including those required for authentication and session continuity) cannot be disabled while continuing to use the Service.

2. How We Use Your Information

A. Provide & Personalize the Service

  • Process uploaded documents using AI to extract structured profile data and seed your Talent Graph.
  • Power Azmuth's conversational profile-building experience (Builder phase before publish; Engagement phase after publish).
  • Run the deterministic JD Mentor scoring pipeline against a frozen snapshot of your profile when you submit a job description.
  • Generate STAR-format resumes (PDF / LaTeX) from your profile, on demand.
  • Draft recruiter-facing outreach copy through the Share feature.
  • Serve your public profile at kanvis.me/<username> and power the AI Twin chat widget that answers visitor questions about you using only your published profile.
  • Personalize your dashboard, completeness metrics, and onboarding progression.

B. Improve & Develop Our Technology

  • Train, retrain, fine-tune, and evaluate our AI agents (Azmuth, JD Mentor, STAR Resume, Share, AI Twin), prompts, tool routing, and orchestration logic using data generated through the Service.
  • Analyze interaction patterns to improve conversational accuracy, JD-fit scoring quality, resume quality, and outreach effectiveness.
  • Develop new features, products, and services informed by aggregated usage patterns and user behavior, including future recruiter-side surfaces (currently waitlist-only).
  • Benchmark and validate the deterministic components of JD Mentor (matchers, scorer) and the LLM components separately to reduce errors and detect regressions.

C. Research & Analytics

  • Generate aggregated, de-identified insights about professional trends, labor markets, skill distributions, and career patterns.
  • Conduct internal research to better understand how candidates describe their experience and how AI can improve career storytelling.
  • Produce anonymized datasets for product development, benchmarking, and commercial purposes.

D. Operations & Security

  • Monitor for abuse, fraud, prompt-injection attempts, scraping, and security threats.
  • Debug, diagnose, and resolve technical issues, including incidents where AI processing fails mid-stream.
  • Run scheduled background jobs (via Google Cloud Tasks) for long-running operations such as JD analysis and resume regeneration.
  • Communicate with you about your account, security alerts, service updates, and policy changes (email delivery via Resend).

3. AI Processing & Third-Party Providers

Operating the Service requires sending portions of your data to third-party providers. We share the minimum data necessary for each provider's role and select reputable vendors with appropriate security practices, but you acknowledge that the following data flows occur:

A. AI Model Providers

Your inputs — including chat messages, uploaded resume content, pasted/uploaded job descriptions, and your profile data — are transmitted to large language model providers (currently including Google Gemini) for:

  • Parsing and enriching your resume into structured profile data.
  • Powering Azmuth's conversational responses and tool decisions during the Build flow.
  • Running the JD Mentor matching pipeline (six parallel matcher LLM calls plus a narrative-generation call per analysis).
  • Generating STAR-format resume content (one LLM call per generation; recompiles use templating only).
  • Drafting recruiter outreach copy in the Share feature.
  • Powering AI Twin responses on your public profile.

Authentication tokens are never included in prompts; they are passed in request bodies and stored in protected context variables during tool execution, so AI models do not see your credentials.

B. Document Parsing Providers

Resumes and job descriptions you upload may be processed by document-parsing providers (e.g., LlamaParse) to convert PDFs and DOCX files into structured text and layout data before LLM enrichment.

C. Authentication

User accounts and sessions are managed through Firebase Authentication (Google Cloud / Firebase). Anonymous sessions, Google Sign-In, and email-code verification are all handled through Firebase.

D. Cloud Infrastructure

The Service runs on Google Cloud Platform in the us-central1 region. This includes Cloud Run (backend and AI service), Cloud SQL (Postgres), Cloud Tasks (background job dispatch), and Google Cloud Storage (file uploads). Files in Cloud Storage are served via signed URLs with limited TTLs. The frontend is hosted on Vercel.

E. Email Delivery

Transactional email (verification codes, account-related notifications) is sent through Resend.

F. Product Analytics and Error Capture

Anonymous and authenticated product analytics, plus client-side exception capture, are sent to PostHog.

G. Responsibility

We select reputable providers and implement appropriate safeguards, but Kanvis is not responsible for the independent data practices of third-party providers. We encourage you to review their privacy policies. A change in our third-party providers does not, on its own, constitute a material change to this Privacy Policy.

4. Data Sharing & Disclosure

We do not sell your personal information. We share information in the following circumstances:

  • Public Profiles: Once you publish, your profile content is publicly accessible at kanvis.me/<username> and may be indexed by search engines, cached by third parties, and shared by visitors. The AI Twin chat widget is publicly reachable through that URL and answers visitor questions using only your published profile.
  • AI Processing Providers: As described in Section 3, the providers we use to deliver the Service receive the inputs they need to do their job.
  • Future Recruiter Features: Recruiter-facing features are currently waitlist-only. If and when they launch, your published profile data may be surfaced to recruiters through search, ranking, or outreach features. Recruiter-side terms and obligations will be published before any such features go live.
  • Service Providers and Subprocessors: Vendors and contractors who help us operate the Service (such as infrastructure, parsing, analytics, and email providers), subject to confidentiality obligations.
  • Aggregated/De-identified Data: We may share anonymized, aggregated, or de-identified data with third parties for research, analytics, industry reports, and commercial purposes. This data cannot reasonably be used to identify you.
  • Legal Obligations: We may disclose information if required by law, regulation, legal process, or governmental request.
  • Business Transfers: In connection with a merger, acquisition, reorganization, or sale of assets, your information may be transferred as part of the transaction. We will provide notice (e.g., via the Service or email) before your data becomes subject to a different privacy policy.
  • Protection of Rights: We may disclose information to protect the rights, safety, and property of Kanvis, our users, or the public — including in response to prompt-injection attacks, scraping, or abuse.

5. Data Retention

  • Active Accounts: We retain your data for as long as your account is active and as reasonably necessary to provide the Service and fulfill the purposes described in this policy. The chat ledger (ChatMessage) is immutable and retained as an audit trail of how your Talent Graph was constructed.
  • Anonymous Sessions: Data from unauthenticated sessions may be retained on our servers. If no account is ever claimed, this data may be anonymized and incorporated into our training datasets, retained for analytics, or deleted at our discretion.
  • Generated Artifacts: STAR-format resumes, JD analyses, and outreach drafts are retained while your account is active so you can revisit, regenerate, and compare them. Deleting an individual artifact through the Service removes it from your account; aggregated and de-identified derivations may persist (see below).
  • Profile Views and Referrals: Coarse view and referral records are retained for your dashboard analytics.
  • Deleted Accounts: Upon account deletion, we will remove your published profile from kanvis.me/<username> and your personal data from active systems within a reasonable timeframe. Note:
    • Search-engine caches and third-party archives are outside our control and may persist.
    • De-identified and aggregated data derived from your usage may be retained indefinitely for AI model improvement, research, and analytics.
  • Backups & Archives: Residual copies of your data may persist in backup systems for a limited period as part of our standard data-management practices.
  • AI Models: You acknowledge that once your data has been used to train or improve AI models, it is not feasible to remove your individual contribution from those models. Trained model weights are not considered personal data.

6. Your Rights & Choices

Depending on your jurisdiction (e.g., the EU/EEA, UK, California), you may have some or all of the following rights:

  • Access: Request a copy of the personal data we hold about you.
  • Correction: Request correction of inaccurate or incomplete personal data. Most of your profile data is also editable directly in product (through Azmuth chat or manual edits).
  • Deletion: Request deletion of your account and personal data, subject to the retention provisions in Section 5.
  • Data Portability: Request your data in a structured, commonly used format.
  • Withdraw Consent: Withdraw consent for specific processing activities, where consent is the legal basis. Note that withdrawing consent does not affect the lawfulness of processing conducted prior to withdrawal.
  • Object to Processing / Restrict Processing: Object to or restrict certain processing in jurisdictions that recognize these rights.
  • Lodge a Complaint: Lodge a complaint with your local data-protection authority.

To exercise any of these rights, contact us at privacy@kanvis.me. We will respond within a reasonable timeframe and in accordance with applicable law.

Cookie Consent: You can update your cookie preferences at any time through the cookie-settings modal in the footer of the Service.

Please note: Certain processing activities (including AI processing of your inputs to deliver the Service and platform-wide model training) are essential to the Service and cannot be selectively opted out of while maintaining an active account.

7. Data Security

We implement industry-standard technical and organizational measures to protect your information, including:

  • Encryption of data in transit (HTTPS/TLS) and at rest where supported by our cloud providers.
  • Firebase-token-based authentication on every authenticated request, with cross-user authorization enforced at the queryset level (you can only ever query your own data).
  • Internal-only endpoints (cross-service traffic between the backend and AI service) protected by an internal-key header and, for background workers, by OIDC token verification.
  • Restricted browser referrers on our Web API key.
  • Connection pooling and rate limiting to defend against scraping and resource exhaustion.
  • Logging, monitoring, and exception capture to detect and respond to anomalies.

No method of transmission or storage is completely secure, and we cannot guarantee absolute security. In the event of a security incident affecting your personal data, we will notify you and the appropriate authorities as required by applicable law.

8. International Data Transfers

The Service is operated from infrastructure in the United States (Google Cloud us-central1 plus Vercel). Your data is transferred to, stored, and processed in the United States and in any other country where our third-party providers operate. By using the Service, you consent to the transfer of your information to these jurisdictions, which may have different data-protection laws than your home country.

For users in the EU/EEA and UK, we rely on appropriate transfer mechanisms (such as Standard Contractual Clauses) where required to lawfully transfer personal data outside your home region.

9. Children's Privacy

The Service is not intended for individuals under the age of 16. We do not knowingly collect personal information from children under 16. If we learn that we have collected data from a child under 16, we will take steps to delete that information. If you believe a minor has provided us with personal information, please contact privacy@kanvis.me.

10. AI-Specific Disclosures

Because the Service is built around AI, we want to be especially clear about a few things:

  • Hallucinations. Large language models can produce confident-sounding output that is factually wrong. Always review AI-generated profile entries, JD analyses, resume content, and outreach copy before you rely on them. See Sections 4(D) and 12 of our Terms of Use.
  • Audit Trail. Every change Azmuth makes to your Talent Graph is linked to the chat message that produced it. You can inspect this evidence trail through the product.
  • AI Twin Constraints. The AI Twin on your public profile is configured with zero mutation tools — it can read your published profile but cannot edit it, send messages on your behalf, or perform actions outside its read-only scope, even if a visitor attempts to prompt-inject it.
  • JD Mentor Determinism. The JD Mentor score (0–100) and its five-component breakdown are computed by deterministic code over LLM-extracted booleans and categorical labels, not by an LLM directly assigning numbers. Re-running an analysis against the same frozen profile snapshot and the same JD will produce the same score. The accompanying narrative (strengths, gaps, action items) is generated by an LLM and may vary.

11. Changes to This Policy

We may update this Privacy Policy from time to time. When we make material changes, we will update the "Effective Date" at the top of this page and may provide additional notice through the Service or by email. Your continued use of the Service after changes are posted constitutes acceptance of the updated policy.

12. Contact Us

If you have any questions, concerns, or requests regarding this Privacy Policy, please contact us at:

Email: privacy@kanvis.me