AI transparency
Institutions reviewing educational software rightly ask hard questions about AI. This page answers them directly, product by product. The short version: at runtime, no student input or student personal information is ever sent to any AI model.
Where AI is — and isn't — used
| Product | AI at runtime (while students use it) | AI in production (behind the scenes) |
|---|---|---|
| Learnova Casebook | None. The story, theory panels and glossary are fixed, human-authored content. | Authors use AI tools as a drafting aid during offline writing; every scene passes a human editorial and quality-control checklist before release. |
| Learnova Signals | None. The game only loads pre-published weekly content files; student forecasts are scored by deterministic rules. | A weekly pipeline rewrites licensed/public news into game dialogue using a commercial large-language-model API (enterprise tier). The pipeline processes news text only — it never sees student data. Output is machine-validated and versioned before publication. |
| CV Studio | None. Career matching, phrasing suggestions and CV health checks run on local rule engines and curated corpora, entirely in the browser. CV content is never sent to an AI model. | Template and corpus production is assisted by AI tools offline, with human review. |
Our commitments
- No student data to models. Student names, emails, inputs and records are never included in any AI request, in any product.
- Paid, enterprise-tier APIs only. Where our production pipeline uses AI, we use commercial API tiers whose terms state that customer data is not used to train models (see the providers' official data-use policies: Anthropic, OpenAI, Google). We never use free or consumer AI tools for any platform content.
- Human in the loop. All AI-assisted content passes automated validation and human review before students can see it.
- Disclosure. If a future feature ever sends student input to a model (for example, an optional AI tutor), it will be off by default, disclosed in advance, gated behind institutional agreement, and engineered so the model receives pseudonymised game-state only — never identity.
Data flow, in one paragraph
Students play in the browser. Game progress is stored locally and (when signed in) in our database, keyed to an institutional account with minimal profile data. Our weekly content pipeline runs on servers, fetches public news, rewrites it with an enterprise AI API, validates the result, and publishes static content files. These two flows never cross: the pipeline has no access to student data, and the games make no AI calls.