AIQ MOOC Engine
RAG-Powered Dynamic Question Generation
🏆 Presented at Int'l Conference on Universities & AI
Parameter-driven RAG architecture that transforms video lecture transcripts into validated, high-quality multiple-choice questions.
What it does
- Designed a parameter-driven system using Llama 3-70B few-shot prompting to automatically extract concept boundaries and generate questions from MOOC video transcripts.
- Integrated Whisper for video-to-text extraction coupled with a vector database retrieval pipeline for grounded context generation.
- Evaluated prompting strategies using RQUGE metrics and LLM-as-a-judge validation, achieving an RAG quality score of 4.46/5 and 83.3% factual accuracy.