Counterfactual Fairness Audits of Multi-Step Clinical LLM Agents Require a Measured Per-Action Instability Floor
研究解决了临床代理在反事实公平性审计中因操作不稳定性导致的结果解读问题，通过测量每项操作的基础不稳定率来解决。
0 citationsRead paper
Selected work
Representative Papers
Seeing Justice Clearly: Handwritten Legal Document Translation with OCR and Vision-Language Models
Low accuracy and fragmented processing pipelines hinder handwritten Marathi legal document recognition and translation—such as First Information Reports (FIRs) and charge sheets—in resource-constrained judicial settings. Method: This paper proposes a lightweight, end-to-end direct translation paradigm tailored to the judicial domain, systematically benchmarking OCR–machine translation (OCR-MT) pipelines against multimodal vision-language models (LLaVA, Pix2Struct) for handwritten Marathi legal document translation. The approach integrates domain-adaptive fine-tuning and a newly curated, handwritten Marathi legal text dataset. Contribution/Results: Experimental results show that the proposed method achieves a +12.3 BLEU score improvement over conventional OCR-MT baselines, reduces inference latency by 40%, and enables offline edge deployment. These advances significantly accelerate digital documentation in Indian grassroots courts, enhancing accessibility and operational efficiency in low-resource legal environments.
0 citationsRead paper
Information availability in different languages and various technological constraints related to multilinguism on the Internet
The dominance of English on the internet imposes significant linguistic barriers and information inequities on non-native English speakers. This study systematically identifies the technical roots of multilingual information imbalance through literature analysis, multi-source data statistics, and assessment of technological trends—quantifying structural underrepresentation of non-English languages across web content, search engines, and AI services, and its impact on global information access. It proposes the “Multilingual Development Framework for Inclusive Interconnection,” an original theoretical contribution addressing gaps in multilingual internet governance. The framework comprises three pillars: language resource development, cross-lingual retrieval enhancement, and localization-aware technology adaptation. Findings provide actionable technical pathways and evidence-based policy recommendations to advance global digital inclusion. (149 words)
0 citationsRead paper
TathyaNyaya and FactLegalLlama: Advancing Factual Judgment Prediction and Explanation in the Indian Legal Context
To address the lack of factual grounding, transparency, and interpretability in AI-based judicial prediction within the Indian legal context, this paper introduces the first “fact-centered” legal AI modeling paradigm. We construct TathyaNyaya—a large-scale, multi-level, structured dataset of judicial fact annotations from Indian case law—currently the largest and most diverse benchmark for Fact-Judgment Prediction and Explanation (FJPE) in India. Building upon it, we propose FactLegalLlama: a joint framework based on instruction-tuned LLaMA-3-8B that performs end-to-end fact-driven judgment prediction and natural language explanation generation. Our approach achieves state-of-the-art performance in both prediction accuracy and explanation relevance and coherence within Indian legal NLP. It is the first to systematically tackle three core challenges: factual anchoring, cross-court generalization, and joint modeling of prediction and interpretability.
0 citationsRead paper
Recent publications
Latest Papers
Counterfactual Fairness Audits of Multi-Step Clinical LLM Agents Require a Measured Per-Action Instability Floor
研究解决了临床代理在反事实公平性审计中因操作不稳定性导致的结果解读问题，通过测量每项操作的基础不稳定率来解决。
0 citationsRead paper
Seeing Justice Clearly: Handwritten Legal Document Translation with OCR and Vision-Language Models
Low accuracy and fragmented processing pipelines hinder handwritten Marathi legal document recognition and translation—such as First Information Reports (FIRs) and charge sheets—in resource-constrained judicial settings. Method: This paper proposes a lightweight, end-to-end direct translation paradigm tailored to the judicial domain, systematically benchmarking OCR–machine translation (OCR-MT) pipelines against multimodal vision-language models (LLaVA, Pix2Struct) for handwritten Marathi legal document translation. The approach integrates domain-adaptive fine-tuning and a newly curated, handwritten Marathi legal text dataset. Contribution/Results: Experimental results show that the proposed method achieves a +12.3 BLEU score improvement over conventional OCR-MT baselines, reduces inference latency by 40%, and enables offline edge deployment. These advances significantly accelerate digital documentation in Indian grassroots courts, enhancing accessibility and operational efficiency in low-resource legal environments.
0 citationsRead paper
Information availability in different languages and various technological constraints related to multilinguism on the Internet
The dominance of English on the internet imposes significant linguistic barriers and information inequities on non-native English speakers. This study systematically identifies the technical roots of multilingual information imbalance through literature analysis, multi-source data statistics, and assessment of technological trends—quantifying structural underrepresentation of non-English languages across web content, search engines, and AI services, and its impact on global information access. It proposes the “Multilingual Development Framework for Inclusive Interconnection,” an original theoretical contribution addressing gaps in multilingual internet governance. The framework comprises three pillars: language resource development, cross-lingual retrieval enhancement, and localization-aware technology adaptation. Findings provide actionable technical pathways and evidence-based policy recommendations to advance global digital inclusion. (149 words)
0 citationsRead paper
TathyaNyaya and FactLegalLlama: Advancing Factual Judgment Prediction and Explanation in the Indian Legal Context
To address the lack of factual grounding, transparency, and interpretability in AI-based judicial prediction within the Indian legal context, this paper introduces the first “fact-centered” legal AI modeling paradigm. We construct TathyaNyaya—a large-scale, multi-level, structured dataset of judicial fact annotations from Indian case law—currently the largest and most diverse benchmark for Fact-Judgment Prediction and Explanation (FJPE) in India. Building upon it, we propose FactLegalLlama: a joint framework based on instruction-tuned LLaMA-3-8B that performs end-to-end fact-driven judgment prediction and natural language explanation generation. Our approach achieves state-of-the-art performance in both prediction accuracy and explanation relevance and coherence within Indian legal NLP. It is the first to systematically tackle three core challenges: factual anchoring, cross-court generalization, and joint modeling of prediction and interpretability.
0 citationsRead paper