Responsible AI
by Design.
Our commitment to student data protection, age-appropriate AI guardrails, transparent algorithms, and pedagogical integrity across the QuestSetu ecosystem and enterprise foundation models.
Six Pillars of Algorithmic Safety
How CodexLabs AI enforces uncompromised pedagogical rigor, child privacy safeguards, and transparent deterministic boundaries across student deployments.
Child Safety & Age-Appropriateness
Multi-layer prompt firewalls, deterministic topic exclusion lists, sentiment monitoring, and mathematical zero-hallucination containment protocols built explicitly for K-12 and higher-ed STEM and literacy curricula.
Privacy & Data Minimisation
We collect strictly pedagogical telemetry. Zero behavioral micro-targeting, zero biometric profiling, zero advertising trackers. All interactions undergo irreversible cryptographic pseudonymization at the boundary proxy.
Dual-Layer Moderation & Oversight
Dual-layer verification combining real-time semantic screening with certified educator review queues. If ambiguity passes threshold, models instantly defer to verified teacher resources rather than generating speculative answers.
Explainability & Audit Trails
Parents, school principals, and educators receive human-readable audit trails detailing why QuestSetu recommended a specific learning milestone, remediation diagnostic, or cognitive challenge.
Model Evaluation & Red-Teaming
Continuous daily fuzzing and adversarial simulation against jailbreaks, dialect and code-switched multi-lingual jailbreaks (Hinglish, Gujarati, Tamil, Marathi), prompt extraction, and subtle social engineering probes.
Privacy & Data Handling
We design data handling around minimisation, access control, retention limits, and applicable Indian privacy requirements. Deployment details are reviewed for each institutional context.
How Every QuestSetu Query is Protected
Observe the automated 4-tier filtering layer executing between learner input and foundational model inference.
Input Firewall
Real-time regex, semantic jailbreak detection, and age-appropriateness screening. PII (names, phone, location) stripped locally prior to network transmission.
Pedagogical Bounds
Context anchoring using localized curricula (CBSE, ICSE, State Boards). System prompts restrict models from generating irrelevant adult themes or non-factual claims.
Output Scrutiny
Secondary discriminator model analyzes generated token streams for bias, tone, toxicity, and educational efficacy before rendering to the student's device.
Audit Telemetry
Anonymized metadata logged to the institutional dashboard for teacher transparency. No student profiling; immutable audit trails available to authorized schools.
Responsible AI Reference Matrix
| FRAMEWORK / REGULATION | JURISDICTION | CodexLabs AI IMPLEMENTATION | COMPLIANCE STATUS |
|---|---|---|---|
| gavel Digital Personal Data Protection Act (DPDP 2023/2026) | Republic of India | Verifiable parental consent for minors under 18; no tracking, targeted advertising, or behavioral micro-profiling of children. | Design Consideration |
| verified Child privacy principles | United States / Global | Zero collection of personal contact data from children under 13 without direct school institutional authorization. | Reference Framework |
| public UNESCO Recommendation on Ethics of AI | International | Human-in-the-loop governance, pedagogical agency preservation, non-discrimination in vernacular languages, and fairness index. | Reference Framework |
| verified_user Security & AI governance principles | Global Standards | We use documented review practices, testing, and data-lifecycle controls appropriate to the current product stage. | Under Review |
Human Oversight & Pedagogical Review
CodexLabs AI treats human oversight as essential. Educational experiences are reviewed through product, safety, and pedagogical perspectives before broader deployment.
Educator Perspective
Educator feedback can inform cognitive scaffolding, classroom suitability, and assessment design.
Child-centred Perspective
Child-centred experience principles guide interaction patterns, healthy engagement, and constructive feedback.
Applied AI Ethics Fellows
Responsible AI review considers model behavior, interpretability, safety boundaries, and data provenance.
Privacy Review
Privacy questions are reviewed against applicable Indian law and user rights.
Incident Response & Reporting
We welcome collaboration from ethical security researchers, schools, and parents. We maintain a zero-retaliation, transparent safe-harbor vulnerability policy.
Detailed Submission
Send findings to our official contact address with reproduction steps, sample prompts, and relevant screenshots or responses.
Initial Review
We review submitted reports, assess their relevance and severity, and prioritize issues that could affect child safety or user privacy.
Remediation & Recognition
Validated issues are routed to the relevant product team for remediation. Contributor recognition is discussed where appropriate.
Are you an Educator or Parent?
If you encountered unexpected or confusing AI responses within QuestSetu, log a direct review ticket with our safety leads.
Discuss Trust & Safety with Our Team
Schools and institutions can contact us to discuss safety principles, privacy expectations, deployment context, and available documentation.