KanAgriIntent-7200: A Multi-Script Kannada Agricultural Intent Classification Dataset

KanAgriIntent-7200 is the first intent classification dataset for Kannada agricultural dialogue, comprising 7,200 manually annotated utterances across 18 intent classes and 105 granular sub-intent labels. The corpus covers three systematically stratified script variants: native Kannada script, code-mixed Kannada-English, and romanised transliterated text. Seven baseline models are benchmarked including classical TF-IDF pipelines and multilingual transformer encoders. Intent-level inter-annotator agreement is κ = 0.9250. This dataset supports research in low-resource NLP, agricultural dialogue systems, and Dravidian language processing.