Research

What I am interested in?

Advised by Ziang Xiao. My research focuses on building and evaluating LLM-powered conversational search agents to enrich information flows across languages and domains. I conduct human subject studies to understand users’ information foraging behaviors, evaluate multilingual and multimodal LLMs in complex search tasks, and develop AI-driven solutions to mitigate information disparity, echo chambers, and overload. My goal is to enhance equitable and effective knowledge access at scale

Research Threads

Propagation and Mitigation of Echo Chambers (ISLE Lab)

Selective information exposure and the formation of echo chambers has been a well-researched topic however we still currently live in a world that is more polarized and with stronger echo chambers. Coming of Generative AI has been dramatic, to say the least, it has been already integrated deeply and rapidly utilized by many organizations across domains. I ask the question, how echo chambers be formed by these LLM-powered applications, and how they can be a blessing in disguise to tackle the currently nearly unsolvable problem of echo chambers?

Multilingual and Cross-lingual understanding (Multilingual NLP Lab)

Reasoning and understanding are crucial topics in NLP which can enable the model to interact with data in ways previously not possible, to ensure that no culture is left behind and we can reason over facts over the texts in the native languages where we can find the most accurate information, we need to create systems that can reason and understand multiple languages at once. I ask the question, how we can create new Natural Language Understanding benchmarks and models that are multilingual?

Surgery Copilots (Intelligence Amplification Lab)

An area where AI can have the most impact on human lives is in the medical field. Recent progress has been made on segmentation and classification tasks in the medical field however a venue that has yet to be tackled is Medical Narratives. However, this is a challenging task because:

current evaluation metrics are not able to capture the depth of generations
this will always be a low-resource task due to data scarcity and changing tools in surgery
the reasoning capabilities of LLMs required for such tasks are not where they need to be.

I am looking to create new methods and metrics to tackle this task.

Understanding Humans at Scale (ISLE Lab)

One of my personal interests and a foundational stone to my research is understanding of humans at scale along with their digital self. Specifically looking at information seeking and foraging with AI agents.

Page updated

Google Sites

Report abuse