hello!

I’m Prabhav Singh (pronounced Pruh-bhav).

I’m currently pursuing my Master’s in Computer Science (Thesis) with a specialization in Human Language Technologies at Johns Hopkins University, where I conduct research at the Center for Language and Speech Processing (CLSP). I’m fortunate to be advised by Prof. Jason Eisner and Prof. Jesus Villalba.

Before this, I earned my Bachelor’s in Electrical Engineering from Delhi University, where I worked with Prof. K.P.S. Rana and Prof. Vineet Kumar at the APC Lab, NSIT.

You can find more details in my CV or read more about me here. Feel free to reach out at: psingh54 at jhu dot edu

Research Interests

My research interests are focused on method development in NLP, with an emphasis on approaches that are adaptable to supervision constraints and aligned with how humans naturally teach, label, and reason. I am interested in developing methods in the fuzzy area, where we must learn from partial feedback, conflicting signals, and implicit preferences.

  1. Clarification, Uncertainty & Attribution: Studying how models navigate ambiguous interactions: knowing when and how to seek clarification, expressing uncertainty and sunderspecification.

  2. Human-AI Collaboration: Building frameworks that optimize decisions in human-LLM workflows, determining when human judgment is needed versus when LLMs suffice (See this and this).

  3. Reasoning in LLMs: Understanding and surfacing uncertainty in LLM reasoning, and grounding reasoning chains in relevant documents from the model’s training data.

Previously, owing to my background in ECE, I have also worked a lot on speech representation and speaker systems. These days, I find my interest more in language and methoids. You can read more about my previous research below:

Previous Interests

I began my research journey with emotion recognition, and while I’ve developed a fair amount of expertise in that space. I was also drawn to speaker recognition and diarization. I found diarization particularly interesting — it's a fundamental speech task with open challenges in temporal structure, multimodal fusion, and low-resource adaptation. I also dabbled in mulimodality: Fusing audio, text, and vision to solve tasks that are natural for humans — but hard for machines. Some of my papers in this field are: