Parminder Bhatia

Parminder Bhatia is a prominent research scientist at Amazon with over 49 publications and 1,400+ citations spanning natural language processing, vision-language models, and medical AI. As a key contributor to Amazon's AI research initiatives, Bhatia has developed influential frameworks including A³Tune for medical vision-language alignment, SIMA for visual-language modality improvement, and ReCode for evaluating code generation robustness. Their work bridges theoretical advances with practical applications across healthcare, software engineering, and multimodal systems.

Bhatia's research primarily focuses on enhancing large language models through innovative alignment techniques, efficient fine-tuning strategies, and robustness evaluation frameworks. Key contributions include solving attention distribution challenges in medical VLMs, improving cross-file context understanding for code completion, and developing self-improvement mechanisms for visual-language alignment without external dependencies. Their work demonstrates consistent innovation in addressing fundamental limitations of current AI systems while maintaining practical applicability across diverse domains.

Analysis of Bhatia's 15 most recent publications reveals a strong emphasis on medical AI applications (40%), code generation/analysis (30%), and foundational LLM improvements (30%). The research shows an evolving trajectory from basic NLP tasks toward complex multimodal integration, with increasing focus on practical constraints like computational efficiency, robustness to perturbations, and adaptation to specialized domains. Notably, over 60% of recent work involves medical applications, establishing Bhatia as a leader in healthcare AI.