Shivam Sharma

I am a Research Assistant at the Ubiquitous Knowledge Processing Lab at TU Darmstadt, working on natural language processing, multimodal understanding, and trustworthy AI systems. My current research focuses on video misinformation detection, video caption verification, and multimodal evidence retrieval for long-form videos.

Before joining TU Darmstadt, I worked at Pacific Northwest National Laboratory on large language models for domain-specific scientific and environmental data. I received an MSc in Data Science from New Jersey Institute of Technology, where my thesis work focused on crisis informatics, and a BTech in Computer and Communication Engineering from The LNM Institute of Information Technology.

Research Focus

• Multimodal misinformation detection: building benchmarks and methods for verifying claims in long-form videos.
• Scientific and environmental NLP: evaluating large language models for domain-specific scientific and regulatory documents.
• Retrieval and long-context modeling: comparing retrieval-augmented systems with long-context language models for evidence-heavy tasks.
• Crisis informatics: modeling social media and multimodal signals during fast-moving crisis events.

Current Work

I am currently developing resources for grounding inconsistencies between video captions and visual evidence. A major goal of this work is to make video verification systems more interpretable: not only predicting whether a claim is wrong, but also identifying where the relevant evidence appears and why the inconsistency matters.