> For the complete documentation index, see [llms.txt](https://mindengage.gitbook.io/mindengage-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://mindengage.gitbook.io/mindengage-docs/troubleshooting-and-faq/faq-generative-ai-course-generation.md). # FAQ — Generative-AI Course Generation MindEngage Creator · Basic vs Enhanced (RAG-powered) modes ### 1. How does MindEngage actually “write” a course? MindEngage uses a large-language-model (LLM) to transform your high-level prompts ( subject, audience, duration ) into: 1. **Outline** — multi-week syllabus 2. **Lesson bodies** — prose, visuals, quizzes, code, simulations 3. **Export/Publish** — HTML site, Google Classroom, etc. In **Standard and Advanced** mode, the LLM relies on its own pretrained knowledge (up to its cut-off date).\ In **Enhanced** mode, we layer a *Retrieval-Augmented Generation (RAG)* step: 1. Your uploaded PDFs/URLs are chunked & semantically indexed. 2. The most relevant chunks are injected into the model prompt. 3. The LLM cites those chunks to increase factual accuracy. > 💡 **RAG = personal library → instant citations → fewer hallucinations.** *** ### 2. Which mode should I pick? | Use-case | Standard | Advanced | Enhanced(LLM + RAG) | | ----------------------------------------------- | ----------- | --------------------- | ------------------------------------------- | | **Rapid brainstorming** | ✅ | Slow Generation | Slow Generation | | **Topic already well-covered in public domain** | ✅ | ✅ | | | **You have proprietary or up-to-date material** | — | - | ✅ | | **Budget-sensitive project** | ✅ (cheaper) | — (≈ 2–3× token cost) | 1.1 to 1.2x token cost compared to advanced | ### 3. How accurate is the generated content? | Scenario | Expected accuracy | | --------------------- | ---------------------------------------------------------------------- | | **Standard (no RAG)** | 80–90 % on mainstream subjects; lower on cutting-edge or niche topics. | | **Advanced (no RAG)** | 80–90 % on mainstream subjects; lower on cutting-edge or niche topics. | | **Enhanced (RAG)** | 90–97 % **for material present in your reference uploads**. | [Google blog for Gemini 2.5 Pro benchmarks ](https://blog.google/technology/google-deepmind/gemini-model-thinking-updates-march-2025) ### 4. How can I improve accuracy? * Upload **clean, authoritative PDFs** (text-selectable, not scanned images). * Tag uploads by topic; pick the most relevant tag when generating. * Keep prompts specific: *“Week 3: Supervised learning – linear regression assumptions”* beats *“Linear regression”*. * Select Enhanced accuracy (LLM + RAG) *** ### 5. Will the AI plagiarise content? * **Basic and Advanced mode** may echo public phrases found across the web. * **Enhanced mode** can quote your reference docs *verbatim*; citations are added automatically. MindEngage generate the citations based on your course settings and you can perform the following: 1. Keep with quotation marks + citation 2. Paraphrase 3. Remove For academic settings run the output through standard plagiarism checkers before publishing. ### 6. What about copyright? | Situation | Are you safe? | Action | | ---------------------------------------------------- | ------------- | --------------------------------------------------------------------------- | | **You uploaded the document & own/hold the licence** | ✅ | Nothing extra. | | **You uploaded third-party material** | ⚠️ | Make sure your licence permits excerpting / derivative works. | | **Basic mode, no RAG** | ✅ | Output is considered new creative text, but still review for long excerpts. | ### 7. Ethical & bias considerations LLMs reflect the biases of their training data. Review for: * **Stereotypes** in examples or analogies * **Cultural imbalance** (Western-centric viewpoints) * **Gendered language** (e.g. “he” for programmers) Use the **Bias Check** button (visible after generation) to highlight flagged sentences. *** ### 8. Are there legal / regulatory issues (FERPA, GDPR, …)? * **Student data** never leaves your browser — MindEngage processes *content*, not learner PII. * Uploaded documents are stored in a GCP bucket with US multi-region. * You can delete all assets you created by deleting the course. *** ### 9. Can I commercialise the generated course? Yes—subject to: 1. **Human review & edits** (the output is “user-generated content”). 2. Respecting any third-party copyright excerpts left in the text. 3. Adhering to the OpenAI and Goolge Gemini usage policies (no disallowed content). *** ### 10. Does the model ever reuse my private uploads for other users? No. The retrieval index is isolated per workspace. The base LLM weights are *not* fine-tuned on your data. *** ### 11. How are citations inserted? In Enhanced mode the model receives each source chunk with a numeric token (`[1]`, `[2]`, …).\ It references the token inline → MindEngage converts to APA / MLA / IEEE in the final HTML. *** ### 12. Why did my job fail at 45 %? * Check the **Usage & Quota** panel — you may have run out of tokens. * PDF may be scanned images ➜ no extractable text ➜ RAG retrieval returned empty. * Retry in **Basic** mode or re-upload a text-recognised (OCR) PDF. *** ### 13. Can I see the prompt that was sent to the model? Yes, during course generation, the detailed course outline is shown and you can further edit it. While generating the visuals and quizzes, the lesson content is used as part of the prompt and MindEngage wraps it further control the output format. *** ### 14. Does MindEngage support non-English courses? * Generation works in **30+ languages** (best in English, Spanish, French, German, Portuguese). * RAG treats Unicode uniformly; mixed-language PDFs are okay. *** ### 15. Where should I ask further questions? * Join our Slack * Email ****