Shubhashis Roy Dipta

Shubhashis is a Ph.D. Researcher in Computer Science at the University of Maryland, Baltimore County (UMBC), advised by Dr. Frank Ferraro. His work bridges Natural Language Processing (NLP) and Computer Vision (CV), with a focus on Decomposition-based Reasoning — enabling machines to reason through complex tasks by breaking them into atomic questions.

His current research explores presupposition-free question-based decomposition for assessing the scientific feasibility of tasks [1]. Previously, he investigated Outcome- and Intention-based Decomposition for video-text retrieval [2], targeting settings where videos lack textual metadata (e.g., social media or surveillance contexts). His earlier work on Hierarchical Variational Autoencoders for Event Representation Learning demonstrated strong applications in text summarization, question answering, and counterfactual reasoning [3].

In Summer 2025, he interned at Amazon (Alexa Team) as an Applied Scientist. He has shown how Deliberative Alignment technique can be used to optimize cost by ruducing token of the tool-calling LLMs (paper coming soon).

In Summer 2024, he interned at Scale AI (Enterprise Team) as a Machine Learning Researcher. He has explored how RL can improve the text2SQL generation by using a small LLM-as-a-Judge during reinforcement loop (Scale AI Blog).

Beyond academia, Shubhashis has a strong background in machine learning systems, applied modeling, and programming. He has ranked among the top 70 on Kaggle 🥉 and placed 8th out of 300+ teams in the ACM ICPC and more. In his initial years, he founded UniShopr (2017–2021) — a cross-border e-commerce platform serving consumers in Bangladesh.