Abstract

This paper explores the integration of advanced NLP techniques—text embeddings, neural networks, and transformer-based LLMs—in the WooScholar App, a tool for natural language search of Independent Study theses. We discuss key methods for word vectorization (e.g., one-hot encoding, Word2Vec, GloVe) and similarity measures (such as cosine distance), and outline neural network architectures and training processes. The study also highlights transformer models that use self-attention for coherent text generation. These concepts are applied in the WooScholar App, built with FastAPI, PostgreSQL, and Next.js, and enhanced by Google’s Gemini API, demonstrating the effective combination of theoretical NLP advancements with practical application development.

Advisor

Visa, Sofia

Department

Computer Science

Disciplines

Other Computer Engineering

Keywords

Text embedding, Neural Networks, Transformer Architecture, Large Language Models, Retrieval-Augmented Generation

Publication Date

2025

Degree Granted

Bachelor of Arts

Document Type

Senior Independent Study Thesis

Share

COinS
 

© Copyright 2025 Soobin Yang