Bridging the AI Divide: Building Explainable, Low-Resource Multilingual LLMs for Inclusive Global Access

This paper presents LegiLLM, a multilingual and explainable large language model framework designed for low-resource Indian languages such as Kannada, Odia, and Assamese. Leveraging fine-tuned transformer models (IndicBERT and XLM-R), we explore explainable AI techniques like LIME and attention heatmaps to provide interpretable reasoning in low-compute environments. Our work aims to bridge the AI accessibility gap for underrepresented communities and support democratized, ethical machine learning across diverse linguistic groups.