Trustworthy and Explainable LLMs for Indian Languages

Indian language AI systems currently suffer from high hallucination rates and poor explainability, largely due to limited training data and complex linguistic structures. This project develops trustworthy and explainable LLMs for Indian languages — building on the BHRAM-IL benchmark to create AI systems capable of generating factually accurate, culturally appropriate, and explainable content.

Objectives

• Design and implement domain-specific hallucination mitigation techniques using RAG and knowledge graph integration
• Create annotated datasets and evaluation metrics tailored for Indian language trustworthiness assessment
• Develop explainability frameworks combining grammar-informed linguistic rules and KG-based justifications
• Build Indian-language-suitable evaluation metrics and conduct systematic validation across diverse languages and domains
• Develop prototype applications including grammar correction assistants and knowledge-guided chatbots

Expected Outcomes

• Open-source datasets for hallucination detection and explainability evaluation across major Indian languages
• Specialized hallucination mitigation and explainability frameworks optimized for Indian linguistic structures
• Comprehensive benchmark suite for trustworthiness in Indian-language LLMs
• Research publications and trained personnel in trustworthy AI methodologies

Funding agency: Anusandhan National Research Foundation (ANRF), Government of India

Scheme: PM Early Career Research Grant (PM ECRG)

Amount: ₹60 Lakhs + overheads

Start Year: 2026

Duration: 3 years

PI: Hrishikesh Terdalkar