KG-Grounded QA for Sanskrit

Building question answering resources from classical Indian knowledge, grounded in an annotated Sanskrit and Ayurveda knowledge graph. The project has two components.

Question Generation develops an automated pipeline that transforms verified knowledge graph triples into natural, domain-faithful Hindi questions for learning, retrieval, and assessment. The work explored LLM-based approaches with ontology-aware prompting before converging on template-based generation for better grammatical precision and terminology preservation.

QA Dataset Construction builds a structured question-answer dataset from Bhavaprakasha Nighantu, preserving domain-critical terminology (Rasa, Guna, Virya, Vipaka) so the resulting dataset is faithful to the source material and useful for downstream NLP and educational applications.

Future directions include multi-hop question generation over the graph and tighter integration with graph-based retrieval.

Team: Triyansh Agrawal, Mudit Guraria