Sentiment Analysis of Nepali COVID-19 Tweets

Built a transformer-based sentiment classification system processing 35K+ Nepali tweets using PyTorch and Hugging Face Transformers, achieving 0.73 F1 score. Developed a preprocessing pipeline for low-resource Nepali language processing and established sentiment analysis benchmarks for pandemic-related social media discourse.

Named Entity Recognition (EverestNER & DanfeNER)

Developed transformer-based NER systems for the low-resource Nepali language using PyTorch and Hugging Face Transformers, achieving F1 scores of 0.85 and 0.80, respectively. Constructed two benchmark datasets: EverestNER (24,587 entities from news articles) and DanfeNER (4,966 entities from tweets) with comprehensive annotation guidelines, establishing the first large-scale NER resources for Nepali language processing.

SMS Spam Detection

Built a binary classification system for 5,574 English SMS messages using Random Forest and Naive Bayes classifiers with NLTK-based feature extraction. Implemented a comprehensive text preprocessing pipeline including tokenization, stopword removal, and feature vectorization using both CountVectorizer and TF-IDF, achieving robust spam detection performance through comparative model evaluation and feature engineering optimization.

E-commerce Project Website

Developed a full-stack web application for the University of Memphis art marketplace using Ruby on Rails backend with MySQL database and a responsive frontend built with HTML, CSS, and Bootstrap. Implemented user authentication, product catalog management, and shopping cart functionality, enabling students and faculty to buy and sell original artworks through a comprehensive e-commerce solution.

Student Answer Assessment in Tutorial Dialogue

Built an ensemble machine learning system for automated short answer evaluation in Intelligent Tutoring Systems using Decision Trees, Random Forest, and Support Vector Regression with NLP feature extraction. Processed hundreds of student-tutor dialogue responses, achieving 80% accuracy in automated assessment, through a scalable ML-driven pipeline.

UofM Local Search Engine

Built an intelligent search platform for the University of Memphis using NLP techniques to process student queries and retrieve relevant campus information links. Implemented search algorithms with text processing and ranking mechanisms to streamline access to university resources through an optimized information retrieval system.