We design language models and intelligent agents that understand context, reason about user intent, and act autonomously in customer-facing environments. Our research explores dialogue management, goal-driven reasoning, and sentiment recognition to create AI systems that can learn from interaction. These systems are designed to perform tasks such as initiating document requests, monitoring case progress, and escalating complex interactions to human specialists, improving the reliability and responsiveness of client communications.

Our research in multimodal reasoning examines how AI can integrate and interpret information from textual, visual, and structural data sources. We develop models that understand complex documents, correlate visual and linguistic elements, and maintain semantic consistency across formats. This enables advanced document automation, including auto-completion of legal forms, cross-validation of supporting materials, and structured data extraction from scanned records, significantly enhancing accuracy and efficiency in compliance-driven environments.

We investigate more efficient alternatives to traditional retrieval-augmented generation, focusing on lightweight knowledge retrieval, context compression, and dynamic memory optimization. Our goal is to build systems that access and apply knowledge efficiently while maintaining transparency and verifiability. These methods support automated summary generation, consistency validation, and contextual data synthesis, enabling scalable and trustworthy automation within digital transformation pipelines.