NLP / low-resource language
Sentiment Analysis of Nepali COVID-19 Tweets
Transformer-based sentiment classification on 35K+ Nepali tweets with PyTorch and Hugging Face Transformers (0.73 F1). Includes a preprocessing pipeline for Nepali text and benchmarks for pandemic-related social media discourse.
NLP / named entity recognition
EverestNER & DanfeNER
NER systems for Nepali using transformers (F1 0.85 / 0.80). Two benchmark datasets: EverestNER (24,587 entities, news) and DanfeNER (4,966 entities, tweets) with annotation guidelines - the first large-scale Nepali NER resources.
Classification / text
SMS Spam Detection
Binary classification on 5,574 English SMS messages with Random Forest and Naive Bayes, NLTK features, and CountVectorizer / TF-IDF for robust spam vs. ham detection.
Full-stack / web
UofM Art Marketplace (E-commerce)
Ruby on Rails, MySQL, and a responsive HTML/CSS/Bootstrap front end. Authentication, catalog, and cart for students and faculty buying and selling artwork.
Intelligent tutoring / assessment
Student Answer Assessment in Tutorial Dialogue
Ensemble models (Decision Trees, Random Forest, SVR) with NLP features for short-answer grading in ITS. Evaluated on hundreds of dialogue responses with ~80% accuracy.
Search / information retrieval
UofM Local Search Engine
Campus-focused search using NLP on student queries, ranking, and link retrieval to streamline access to University of Memphis resources.