AI Policy
Learnvelo, Inc. | Effective Date: March 30, 2026 | Last Updated: June 30, 2026
This policy describes how Learnvelo uses artificial intelligence across its platform. It ensures that AI is used responsibly, transparently, and in service of educational outcomes — with appropriate human oversight at every stage.
As AI plays an expanding role in educational technology, institutional partners and regulators increasingly require vendors to demonstrate deliberate AI governance. This policy provides that assurance and serves as the authoritative reference for how AI decisions are made at Learnvelo.
1. Scope
This policy applies to all AI and machine learning capabilities within the Learnvelo platform, including:
- AI-powered content generation (course pages, reading materials)
- AI-assisted assessment creation (quiz and exercise generation)
- AI mentor chat (student-facing conversational support)
- AI-assisted grading (open-ended response evaluation)
- AI-powered practice activities (math solver, debate arena, and future formative tools)
- Learning recommendations and adaptive pathways
- RAG-enhanced content retrieval (vector search over textbook materials)
- Image generation for course materials (OpenAI image model)
- Video suggestion integration
2. Governing Principles
Learnvelo’s use of AI is guided by six principles:
2.1 Educational Purpose
AI features exist to enhance teaching and learning, not to replace the instructor’s role. Every AI capability must have a clear educational justification — improving content quality, providing timely feedback, personalizing learning paths, or reducing barriers to access.
AI is not used for surveillance, behavioral prediction unrelated to learning, marketing profiling, or any purpose outside the educational mission.
2.2 Human Oversight
Instructors maintain authority over all content and grading decisions that affect student academic records. AI serves as a tool under instructor control, not an autonomous decision-maker for consequential academic outcomes.
- AI-generated content requires instructor review before publication
- AI-assisted grading on summative assessments requires instructor approval
- Instructors can disable, override, or modify any AI feature for their courses
2.3 Transparency
Learnvelo is transparent about its use of AI with all stakeholders:
- Institutions receive full disclosure of AI capabilities, data practices, and governance through this policy and related compliance documents
- Instructors are informed about how AI features work, what data they use, and how to control them
- Students are informed that the platform uses AI features (via the EULA and Privacy Policy), including the AI mentor chat, which is clearly presented as an AI assistant
2.4 Data Minimization
AI features use the minimum data necessary to fulfill their educational purpose. Student personally identifiable information is never sent to AI service providers. See our Privacy Policy for detailed data practices.
2.5 Fairness and Bias Awareness
AI-generated content and assessments must be reviewed for accuracy, bias, and appropriateness before reaching students. Learnvelo recognizes that AI models can reflect biases present in their training data and implements controls to mitigate this risk.
2.6 Continuous Improvement
AI governance is not static. Learnvelo regularly reviews its AI practices, monitors for emerging risks, and updates this policy as AI capabilities evolve and regulatory guidance matures.
3. AI Feature Inventory
3.1 Content Generation
| Feature | Description | Human Oversight |
|---|---|---|
| Page content generation | Generates reading material, explanations, and examples based on module learning objectives and textbook content | Instructor must review and approve before publication |
| Quiz and exercise generation | Creates assessment questions aligned with page content and learning objectives | Instructor reviews generated questions; can edit, delete, or regenerate |
| Image generation | Creates illustrative images for course pages based on text descriptions | Instructor reviews generated images; can replace or remove |
| Video suggestions | Suggests relevant YouTube videos based on page content | Instructor reviews suggestions; videos are not auto-embedded without review |
3.2 Student-Facing AI Features
| Feature | Description | Human Oversight |
|---|---|---|
| AI mentor chat | Conversational learning assistant that answers student questions about current page content | Clearly presented as an AI assistant; scoped to current course content; does not access student records |
| Math solver | AI-guided step-by-step problem solving with hints and feedback | Formative only (practice, not graded for academic record) |
| Debate arena | AI opponent for structured argumentation practice with scoring | Formative only (practice, not graded for academic record) |
3.3 AI-Assisted Grading
| Feature | Description | Human Oversight |
|---|---|---|
| AI-assisted text grading | AI evaluates student text submissions and assigns a score with written feedback. Currently implemented for open-ended responses in assessment quizzes; the same approach applies to any future activity that uses AI to grade student text. | Two instructor-configurable modes per activity. Manual (default): the AI produces a suggested score that is withheld from the student until an instructor reviews and confirms it before it counts. Automatic (instructor opt-in, where an activity offers it): the AI score and feedback are applied and shown to the student immediately — and posted to the institutional LMS for LTI courses — without prior instructor review. In both modes instructors can override any AI-assigned grade, and students may request human review. |
3.4 Learning Recommendations
| Feature | Description | Human Oversight |
|---|---|---|
| Content recommendations | Suggests next content based on aggregated learning patterns | Based on anonymized, aggregated data; no individual profiling |
4. AI Decision Framework
4.1 When AI May Act Autonomously
AI may operate without per-instance human review in the following contexts, where the educational stakes are low and immediate feedback improves learning:
| Context | Justification | Safeguards |
|---|---|---|
| AI mentor chat responses | Real-time conversational support requires immediate responses; content is scoped to course material | Responses limited to current page context; no access to student records; clearly labeled as AI |
| Formative practice activity scoring | Practice exercises require instant feedback to be pedagogically effective; scores do not affect academic records | Clearly framed as practice; not included in gradebook; no impact on course grade |
| Automatic AI grading of student text submissions (opt-in; Manual review is the default) | When an instructor enables Automatic grading for an activity, scores and feedback are applied immediately at scale | AI receives no identifying student data; instructors can return the activity to Manual mode (the default), can override any grade, and students may request human review. This mode affects academic records, including LMS grade passback for LTI courses |
4.2 When AI Requires Human Approval
AI output must be reviewed and approved by an instructor before it affects students or academic records:
| Context | Requirement | Rationale |
|---|---|---|
| Course content publication | Instructor review and approval required | Content accuracy and alignment with learning objectives require expert judgment |
| Assessment creation | Instructor review of all AI-generated questions | Assessment validity and fairness require instructor expertise |
| Summative grading (default) | By default (Manual grading), an instructor must review and confirm each AI-suggested score before it is recorded or shown to the student | Academic integrity and institutional accreditation standards favor human accountability for grades. An instructor may opt an activity into Automatic grading, which applies AI grades without prior review (see Section 5.1) |
4.3 When AI Is Not Used
Learnvelo does not use AI for:
- Admissions or enrollment decisions — enrollment is open or institution-controlled
- Student disciplinary actions — no AI-driven behavioral monitoring or flagging
- Automated academic integrity enforcement — no AI plagiarism detection or cheating accusation
- Surveillance or behavioral profiling — no tracking of student behavior for non-educational purposes
- Marketing or advertising targeting — no AI-driven ad personalization using student data
- Grading that an instructor cannot review or override — instructors can always revise or override any AI-assigned grade, and students may request human review of an automatically assigned grade
5. Data Governance for AI
5.1 Data Sent to AI Providers
AI text features are served by Learnvelo’s configured providers (OpenAI and/or Anthropic) through a provider gateway; the data sent is identical regardless of which provider serves a request. Image generation and embeddings (Pinecone RAG) use OpenAI only.
| Feature | Data Sent | Data NOT Sent |
|---|---|---|
| Content generation | Module title, learning objectives, textbook passages (via RAG) | Student names, emails, grades, enrollment data |
| AI mentor chat | Current page content, student’s question | Student name, email, grade history, other students’ data |
| Quiz generation | Page content, learning objectives | Student performance data |
| AI-assisted grading | Assessment rubric, student’s response text | Student name, email, other identifying information |
| Practice activities | Problem parameters, student’s work-in-progress | Student name, email, grade history |
| Image generation (OpenAI only) | Text description of desired image | No user data of any kind |
5.2 Data Protection Controls
- No student PII in prompts. Enforced at the application level; prompts are built from course content and anonymized inputs
- No model training on Learnvelo data. Under OpenAI’s and Anthropic’s commercial/API terms, API data is not used to train their models. Learnvelo does not store prompt logs. No zero-retention arrangement is currently in place; provider data retention follows each provider’s standard commercial terms
- Vector store contains only textbook content. Our vector database stores embeddings of course reference materials, not student data
- Data Processing Agreements. DPAs are in place with OpenAI, Anthropic, and Pinecone
5.3 Instructor Control Over AI Data
- Opt-out: Instructors can disable AI features for their courses entirely
- Content control: Instructors choose which textbook materials are indexed for RAG
- Grading control: Instructors set each activity’s grading mode — Manual (the default: AI suggests a score that an instructor confirms before it counts) or Automatic, where offered (AI grades and scores immediately) — and can override any AI-assigned grade
- Review authority: All AI-generated content requires instructor approval before publication
6. Content Quality and Accuracy
6.1 Content Review Process
| Stage | Responsibility | Checks |
|---|---|---|
| Generation | Platform (automated) | Content is generated using RAG-enhanced prompts anchored to provided textbook materials to improve factual accuracy |
| Instructor review | Instructor (mandatory) | Accuracy, completeness, appropriateness, alignment with learning objectives, absence of bias or harmful content |
| Publication | Instructor (explicit action) | Instructor affirmatively publishes content, accepting responsibility per EULA Section 4.8 |
6.2 AI Limitations
Learnvelo discloses the following limitations of AI features:
- AI-generated content may contain factual errors, outdated information, or incomplete explanations
- AI responses in the mentor chat are not a substitute for instructor guidance
- AI-assisted grading suggestions may not capture all nuances of student understanding
- AI practice activity scoring is approximate and designed for learning feedback, not summative evaluation
6.3 Source Attribution
When AI-generated content incorporates information from textbook sources, source documents are tracked and stored with the content. Instructors can view which sources informed the generation, supporting academic honesty and allowing verification against original materials.
7. Fairness, Bias, and Safety
7.1 Bias Mitigation
| Risk | Mitigation |
|---|---|
| Biased content generation | RAG anchoring to vetted textbook materials reduces reliance on model’s general training data; instructor review catches bias before publication |
| Biased assessment generation | Instructor review of all generated questions; questions derived from specific course content, not open-ended generation |
| Biased grading | Rubric-based evaluation reduces subjectivity; AI prompts exclude demographic data; instructors can override any AI-assigned grade and can require Manual review (instructor confirmation before a grade counts) for higher-stakes assessments |
| Cultural or linguistic bias | Instructor review is the primary safeguard; AI prompts do not include demographic data that could trigger differential treatment |
7.2 Content Safety
- AI provider content filtering: Built-in safety filters prevent generation of harmful, illegal, or inappropriate content
- Instructor review gate: No AI content reaches students without instructor approval
- Platform content standards: All content must comply with the Acceptable Use Policy and Content Standards in our EULA
8. AI Service Providers
| Provider | Services Used | Safeguards |
|---|---|---|
| OpenAI | Text generation, chat, grading, and activities; image generation; embeddings (Pinecone RAG) | SOC 2 Type II; Data Processing Agreement; API data not used to train models; no zero-retention arrangement currently in place |
| Anthropic | Text generation, chat, grading, and activities on selected model tiers (Claude), routed via Learnvelo’s provider gateway | SOC 2 Type II; Data Processing Agreement (incorporated in Anthropic’s Commercial Terms, includes SCCs); API data not used to train models; no zero-retention arrangement currently in place |
| Pinecone | Vector database for RAG-enhanced content generation | SOC 2 Type II; Data Processing Agreement; stores textbook embeddings only (no PII) |
Before adopting any new AI service provider, Learnvelo evaluates data handling practices, security certifications, availability of a Data Processing Agreement, content safety mechanisms, and educational suitability for student populations.
9. Regulatory Compliance
9.1 FERPA
- No student educational records are sent to AI providers
- By default (Manual grading), AI-suggested grades are instructor-confirmed before entering the educational record; where an instructor opts into Automatic grading, AI grades are applied directly and remain subject to instructor override and student-requested review
- Institutions retain control over their students’ data as data controllers
9.2 GDPR
- AI processing is based on legitimate interest (enhancing educational experience) with minimal data use
- Students can opt out of AI mentor chat
- Instructors can disable AI features for their courses
- By default (Manual grading), AI-suggested grades are confirmed by an instructor before taking effect, so the default grading flow is not solely automated. Where an instructor opts an activity into Automatic grading, a grade may be produced by solely automated means; Learnvelo then provides the safeguards expected under Article 22(3): instructors can override any AI-assigned grade, students may request human review, and the activity can be returned to Manual mode (the default)
9.3 Institutional AI Policies
Learnvelo recognizes that institutions may have their own AI governance policies. To support institutional compliance:
- Instructors can disable AI features per-course to comply with institutional restrictions
- AI-assisted grading is opt-in, allowing institutions to prohibit it if their policies require
- This policy and related data documentation are available for institutional review upon request
- Learnvelo will work with institutions to accommodate specific AI governance requirements within contractual agreements
10. Contact
For questions about this AI policy or Learnvelo’s use of artificial intelligence, contact us at support@learnvelo.com.
For detailed information about how we handle your data, see our Privacy Policy. For the full terms governing use of the Service, see our End User License Agreement.
This policy is reviewed annually and when new AI capabilities are introduced to the platform.