The case for syllabus-grounded AI in medical education

A generic chatbot that is confidently wrong is a nuisance in most contexts. For a medical student preparing for the NMC exams, it is dangerous. ByteAstra takes the opposite approach to a general assistant. It answers only from curated, open-licensed content mapped to official NMC CBME competencies, using a retrieval-augmented generation pipeline with nineteen subject-isolated collections (one per subject) so answers never bleed across disciplines. The result is measurable: 94% accuracy on BAMS university-paper benchmarks, zero hallucination, and a citation on every answer. Here is how the pipeline is built, and why grounding beats scale for exam-relevant tutoring.