My Projects

Presented at Oregon Institute of Technology's annual IdeaFest, a celebration of student scholarship and creativity. AI tools are increasingly used in CS education for code explanation, debugging, and feedback, yet most evaluations hide a critical reality: the same model can answer the same prompt differently across repeated runs, and educational AI must be trustworthy and reproducible, not just performant.

Paper accepted at the American Society for Engineering Education (ASEE) Annual Conference, 2026. As LLMs become routine tools in CS1 courses, their instructional reliability remains under-scrutinized; models that generate correct code may still misdiagnose the conceptual errors novice programmers actually make. This study examines diagnostic alignment, cross-model variability, and prompt sensitivity.

Published at ACL 2026. Language models fail at structured-data reasoning not because they cannot produce SQL, but because they misalign questions with the schema before SQL generation begins. GRIPS is a neurosymbolic framework that separates question-to-schema grounding from SQL generation, improving accuracy by an average of 11 REMS points while reducing token usage by 47.8% across two benchmarks.

This portfolio website, built as the CS 463/563 Final Project. A fully responsive multi-page site using HTML5, CSS3 with custom properties, Bootstrap 5, and vanilla JavaScript (ES6+). Features include scroll animations, Bootstrap Offcanvas mobile nav, form validation, and a typing animation hero.

A responsive weather dashboard that fetches real-time weather data from the OpenWeatherMap API. Features a 5-day forecast, geolocation support, unit toggling (°F / °C), and animated weather icons. Built with vanilla JavaScript and Bootstrap.

A text classification pipeline using BERT and fine-tuned transformers to perform multi-class sentiment analysis on product reviews. Achieved 92% accuracy on the test set. Includes a simple Flask API endpoint for inference.