Harsh Mishra

Research Intern @Dr. B. R. Ambedkar National Institute Of Technology Nit , Jalandhar

Jalandhar, IN

I have led three impactful AI/ML projects addressing challenges in low-resource and domain-specific language settings. In the CMIR (Code-Mixed Information Retrieval) project, I developed a robust retrieval system for Bengali-English Romanized social media data. Using a novel benchmark dataset from Facebook groups and integrating FAISS-based dense retrieval with CrossEncoder reranking, our system significantly improved Mean Average Precision (MAP), enabling efficient access to local information in informal, multilingual environments.The Varanasi Tourism Hindi QA project focused on building a culturally-aware question-answering system for Hindi-speaking tourists. By curating a domain-specific dataset around Varanasi\'s temples, ghats, and rituals, and deploying a context-sensitive NLP model, the system provides accurate responses in Hindi—bridging a crucial gap in regional digital tourism solutions.In SCHIHigh, I proposed a hybrid scientific document summarization framework combining PEGASUS, BART, and Longformer (LED) models. This ensemble approach enhances semantic understanding, fluency, and long-context processing. Evaluated using ROUGE metrics, it outperforms standalone models in generating coherent, informative highlights from scientific texts.Collectively, these projects showcase my focus on inclusive NLP, information access in underrepresented languages, and domain-aligned AI solutions with real-world impact.

GLA University

Bachelor of Technology - BTech