cv

AI/ML Intern

US Department of Agriculture

Applied machine learning techniques in microbiome dataset to identify growth-enhancing fungi in swine.

• Machine learning
• Microbiome
• Pattern identification
• Swine growth

Graduate Teaching Assistant

North Carolina State University

Assist students with homework assignments, projects, and exams. Conduct office hours. Prepare learning materials. Provide feedback on slides, presentation skills, and reports.

• Discrete Mathematics
• C and Software Tools
• Artificial Intelligence (AI)
• Technical Communication

Graduate Research Assistant

North Carolina State University

Researching on estimating creativity using psychological properties in image domain.

• Creativity
• Psychology
• Image understanding

Master's in Computer Science

North Carolina State University

Machine learning and Creativity

• Data Structures and Algorithms
• Artificial Intelligence
• Data Mining
• Neural Networks
• Efficient Deep Learning
• Software Engineering
• Accelerated Deep Learning
• Advanced Machine Learning
• Technical Communication

PhD in Computer Science

North Carolina State University

Machine learning and Creativity

• Dissertation title: Modeling Creativity Dimensions to estimate Image Creativity

Title winner at Games and Entertainment

LOCUS Festival, IOE, Pulchowk Campus

Undergrad project

Machine learning models reveal Saccharomyces yeasts are associated with poor piglet growth in early development.

Journal of Animal Science

PromptIQ: Who Cares About Prompts? Let System Handle It – A Component-Aware Framework for T2I Generation.

arXiv preprint arXiv:2505.06467

A metaproteomic analysis of the piglet fecal microbiome during the weaning transition.

Frontiers in Microbiology

Automated image generation via prompt-based guidance

Developed a fully automated user-driven T2I framework using prompt engineering; discovered major limitations in diffusion-based models for complex object generation. Introduced a novel CAS metric and achieved visual flaw detection accuracy by nearly 240% (0.54 vs. 0.16) over the CLIP baseline across four vehicle categories.

• text-to-image
• prompt engineering

Few-shot Learning for Energy Detection

Developed a few-shot learning model using Prototypical Networks and RepPoints embeddings for aerial imagery classification. This model achieved 95% accuracy on novel energy facility classes with limited training data.

• few-shot learning
• prototypical networks
• image embeddings

Evaluation of various BERT algorithms

Systematically benchmarked five BERT model variants across eight GLUE tasks, rigorously identifying the optimal performer for Natural Language Understanding (NLU). The analysis revealed key performance differences between semantic and syntactic tasks.

• BERT algorithms
• GLUE tasks
• natural language understanding

EcoNet Weather Forecasting

Built an ensemble classification model using SMOTE and CatBoost on a 6.5 million-sample, severely imbalanced dataset. This system achieved 99% recall, ranking 1st among 25 competing teams.

• machine learning
• class imbalance
• classification

Pneumonia Detection

Applied the DenseNet deep learning architecture to chest X-ray data for automated pneumonia detection and localization. This project demonstrated proficiency in medical imaging analysis by achieving 80% accuracy on the diagnostic classification task.

• deep learning
• computer vision
• medical image analysis