Dishcovery
Dishcovery started from a simple truth: I love food, and I’m lucky enough to be an adventurous eater with zero dietary restrictions. But I quickly learned that coordinating meals …
Professional Summary
I’m Dhruv Madhwal, a CS grad student at Arizona State University who likes building software at the intersection of ML, NLP, and reliable AI. My work focuses on hallucination detection, multi-hop question answering, LLM/VLM evaluation, and compositional reasoning in vision-language models, including research published at ACL. I enjoy turning research ideas into working software: agents, retrieval pipelines, ML applications, backend services, and tools people can actually use. I also spend time on the systems side: event-driven architectures, distributed data platforms, scalable pipelines, and production-oriented engineering.
Education
MS Computer Science
2024-08-22
2026-08-01
Arizona State University
MS Physics
2017-08-01
2022-05-31
Birla Institute of Technology and Science, Goa Campus
BE Electronics
2017-08-01
2022-05-31
Birla Institute of Technology and Science, Goa Campus
Interests
Large Language Models
Agenic AI
Computer Vision
NLP
Software Engineering
Distributed Systems
Projects
I enjoy making things. Here are a selection of projects that I have worked on over the years.
Multi-Hop Reasoning Agent
Large language models can answer many questions directly, but they often struggle when a question requires several connected steps. Research shows that breaking complex questions …
Machine Unlearning in Small Language Models
Large language models can memorize facts during training, but removing a specific piece of knowledge after training is not straightforward. Retraining a model from scratch is …
📚 My Research
I’m a thesis student at Arizona State University working on reliable AI for language models. My thesis studies how to detect hallucinations and knowledge gaps in LLMs for multi-hop question answering, with the goal of making models better at knowing when they should abstain.
I have also worked on CLIP-style vision-language models and compositional reasoning, and have built NL-to-SQL benchmarks for privacy-sensitive domains such as healthcare, law, and criminal justice.
Please reach out if you’re interested in collaborating on reliable AI, LLM/VLM evaluation, or applied ML systems.
Publications
Decomposed Prompting Does Not Fix Knowledge Gaps, But Helps Models Say "I Don't Know"
Decomposed prompting does not reliably repair missing knowledge in closed-book QA, but disagreement across prompting regimes provides a strong signal for when models should abstain …
Concrete Jungle: Towards Concreteness Paved Contrastive Negative Mining for Compositional Understanding
We study how concreteness-aware negative mining can improve compositional understanding in vision-language models, and introduce ConcretePlant, Cement loss, and Slipform to make …
eun-woo-im