Multimodal and Multi-Turn Intent Recognition
LLM-based intent recognition across multimodal inputs and extended dialogue contexts.
Research Scope
During a research internship at Lenovo Research Institute, I worked on LLM-based multimodal and multi-turn intent recognition. The project examined how a system can interpret user intent when relevant evidence is distributed across different modalities or multiple dialogue turns.
Focus
My work included reasoning-strategy selection, model training, and systematic evaluation. The broader goal was to improve how language-model-based systems combine contextual evidence and select an appropriate reasoning process for different interaction conditions.