Research Areas
Advancing Responsible, Scalable, and Interpretable AI across Core ML, Law, and Healthcare
Research Philosophy
My research philosophy centers on building AI systems that are not only accurate but also transparent, accountable, and socially beneficial. I believe that the next generation of AI for high-stakes domains like law and healthcare must prioritize interpretability, factual grounding, and fairness. By combining advances in natural language processing, large language models, and domain expertise, I aim to create tools that augment human decision-making rather than replace it.
Legal AI
Building AI systems for legal judgment prediction, explanation, retrieval, rhetorical role segmentation, and document generation in the Indian and global legal contexts.
Core AI/ML
Advancing foundational AI and machine learning techniques including model architectures, optimization, and learning paradigms for next-generation intelligent systems.
Evaluation & Benchmarking
Designing rigorous evaluation frameworks, benchmarks, and metrics to assess AI system performance, fairness, and reliability across diverse tasks.
Explainable AI
Developing transparent and interpretable AI systems that provide human-understandable reasoning, mechanistic interpretability, and justification for predictions.
Multilingual AI
Developing AI technologies that work across Arabic, English, Indic languages, and other multilingual settings with cross-lingual transfer capabilities.
LLM Reasoning & Agents
Exploring reasoning capabilities, agentic workflows, tool use, and multi-step problem solving in large language models for complex domain tasks.
Healthcare AI
Creating AI systems for medical dialogue, clinical decision support, predictive modeling, and patient-centric care across multilingual settings.
AI Memory Management
Developing token-efficient memory systems for large language models that compress conversation history without losing critical context and enable extended reasoning.
Distributed Training
Researching scalable distributed training paradigms and efficient model parallelism strategies to train large AI systems across heterogeneous compute clusters.
Test-Time Training
Investigating test-time training and adaptation techniques that enable language models to specialize to unseen domains at inference without expensive fine-tuning.