Security & Architecture Overview

In products designed for children, security and privacy cannot be superficial afterthoughts. Meet Anvi is engineered around the principle of Privacy by Design: we minimize what we process, enforce hardware-level isolation, and execute learning intelligence directly on your device rather than in distant cloud servers.

1. Core Architectural Safeguards

2. On-Device Speech Recognition Pipeline (Thinking)

When a child participates in the Thinking activity, their spoken reflection is processed through a strict local sandbox:

• Permission Gate: The microphone is disabled by default. The system permission prompt is only triggered after a parent activates Thinking behind the PIN gate.
• Local Apple Speech Recognizer: The app leverages iOS's native on-device speech engine. Audio streams are buffered in volatile memory only for the duration of the child's response.
• Zero Remote Recording: Audio is discarded immediately after text transcription. No audio files are ever written to disk or transmitted to our servers or third-party speech APIs.
• Local Sandbox Storage: The resulting child drawing and text transcript are stored solely in the app's local device container. They do not sync to your cloud account and cannot be read by other apps.

3. On-Device AI Pipeline (Math Wizard)

Math Wizard provides adaptive guidance without compromising privacy:

• Apple Foundation Models Framework: On supported devices with Apple Intelligence, the system language model generates hint variations locally.
• No Third-Party AI API Calls: Meet Anvi does NOT send child prompts, errors, or session data to external cloud LLM providers (e.g. OpenAI, Anthropic, Google Cloud).
• Deterministic Fallback: If on-device models are unsupported or unavailable, the app relies exclusively on pre-authored, clinically vetted static hint strings.

4. Network Traffic & Backend Boundary

The only outbound network requests generated by the Meet Anvi mobile app are:

• Parent Authentication: Secure password-based or Apple/Google OAuth sign-in to Supabase over HTTPS.
• Learning Progress Synchronization: Syncing non-sensitive progress markers (child profile level, completed lesson stops, stars earned, and streak days) so children can learn across shared family devices.
• Apple StoreKit Receipt Validation: Communicating with Apple's servers to verify active subscription entitlements.

5. Website Security & Analytics

Our marketing website (meetanvi.com) mirrors our app privacy ethos:

• No Third-Party Tracking Pixels: We do not run Meta Pixel, TikTok Pixel, Google Ads remarketing, or third-party cookies.
• Anonymous Vercel Web Analytics: Collects aggregated page views with zero cookies, zero IP storage, and a privacy-preserving daily rotating hash.
• Strict Content Security: Enforced modern HTTPS, DNSSEC, and secure headers on Vercel's global edge network.

6. Access Control & Least Privilege

Administrative access to cloud databases is governed by strict technical controls:

• Mandatory multi-factor authentication (MFA) on all developer and administrator accounts.
• Row-Level Security (RLS) enforced at the PostgreSQL engine level, ensuring users can only read and write their own family records.
• No production database credentials or API secrets are stored inside the mobile client binary or public repositories.

7. Privacy by Design: 10-Question Engineering Checklist

Before any new feature, API route, or SDK integration is merged into the Meet Anvi codebase, engineering leads must answer our mandatory Privacy by Design checklist:

• 1. Necessity: Does this feature strictly require personal data to deliver early learning?
• 2. Minimization: Can this feature function with pseudonymous identifiers or non-identifying telemetry?
• 3. Child Context: Does this touch a child user directly? (If yes, zero personal data may be collected).
• 4. Local Execution: Can processing run 100% on-device (e.g., Apple's on-device speech recognition) without cloud transmission?
• 5. Parental Consent: Is explicit parental authorization required prior to activation?
• 6. Parent Gate: Is the entry point isolated behind our 4-digit PIN gate or mathematical challenge?
• 7. Third-Party Scrub: Does any external SDK or cloud API receive payload data? (Zero third-party SDK policy).
• 8. Retention Limit: What is the exact expiration timeline for temporary cache or records?
• 9. Frictionless Erasure: Is this data automatically wiped when a parent deletes their account or resets progress?
• 10. Policy Transparency: Does this require an update to our public Data Inventory or App Store Nutrition Labels?

8. Data Protection Impact Assessment (DPIA) Summary

In accordance with Article 35 of the EU/UK GDPR and India DPDP guidelines, Meet Anvi carried out an internal privacy risk review of its speech recognition and hint features. This is a summary of that review, not a formal regulatory DPIA:

9. Incident Response & 72-Hour Breach Notification Protocol

We maintain a documented Incident Response Plan to safeguard data integrity and ensure rapid regulatory compliance:

10. Vulnerability Disclosure & Security Contact

We welcome responsible vulnerability disclosures from security researchers and developers. If you identify a potential security vulnerability in Meet Anvi's systems:

• Email our team at: info@meetanvi.com.
• Please include detailed reproduction steps. We acknowledge security reports within 24 hours and commit to rapid remediation.