Projects

ZotFinder Campus Navigation: UX Research & Interface Redesign

Overview: A research-driven redesign of UCI’s ZotFinder app to better support off-campus commuters—from planning safe routes to finding quiet, comfortable study spaces on campus.

What I did:

• Interviewed and observed students to understand pain points in commuting, wayfinding, and study-space selection.

• Created a persona and scenario, then designed an interactive mobile prototype in Figma.

• Ran think-aloud usability tests with student participants and synthesized feedback into concrete UI and interaction improvements.

👉 Visit ZotFinder Redesign Interface

👉 Contextual Inquiry 👉 Sketching and Prototyping 👉 Usability Evaluation

---

Age- and Context-Appropriate AI Chat for Children

Overview: An interactive AI chatbot tailored to three developmental bands—ages 5–7, 8–12, and 13+; built with OpenAI ChatGPT API and Retrieval-Augmented Generation (RAG) for accurate, age-appropriate, context-aware replies.

How it works:

• Collects age and region at conversation start to customize tone, vocabulary, and examples.

• Uses RAG to ground responses in curated knowledge, improving safety and relevance.

• Prioritizes responsible interaction patterns for kids.

👉 Visit A Little Secret

---

Capstone Project — Delayed Antibiotics Administration

Partner: Children’s Hospital of Orange County (CHOC)
Duration: Jan 2025 – Jun 2025
Goal: Predict and classify the severity of delayed antibiotic administration to support proactive clinical intervention.

My contributions:

• Data Engineering: Built an ETL pipeline to consolidate multi-source pediatric data; enabled reliable feature engineering and modeling.

• Model Development: Led a Random Forest classifier; hyperparameter tuning, feature analysis, and baselines against alternative models.

• Evaluation & Communication: Assessed with ROC curves and confusion matrices; presented findings and implications to clinical stakeholders.

• Collaboration & Management: Translated clinician requirements, validated outputs, and maintained code/docs in GitHub.

👉 Project site

---

Breast Cancer Recurrence Prediction

Focus: This project applies Bayesian Logistic Regression to predict five-year breast cancer recurrence using a structured clinical dataset (n = 275). The goal is to build an interpretable, uncertainty-aware modeling workflow that supports clinical decision-making.

Outcome: The Bayesian logistic regression model identified axillary lymph node involvement (3–11 nodes) and higher malignancy degree as strong predictors of recurrence, while small tumors (10–14 mm) were associated with lower risk. Using a ROC-optimized threshold of 0.33, the model achieved ~67% sensitivity, ~71% specificity, and ~70% cross-validated accuracy, providing interpretable, uncertainty-aware risk estimates for breast cancer recurrence.

👉 Detailed report 👉 R code

---

Factors Related to Computer Science Conference Publications

Question: Which variables are associated with higher conference publication output?

Approach:

• Extensive data cleaning/wrangling/mining on open-source datasets.

• Used Python to parse and aggregate JSON data from NSF and NIH, aligning funding/activity features with publication counts.

• Wrote SQL queries to extract and join relational data across multiple tables.

• Built tidy, analysis-ready tables for modeling and comparison.

Outcome: Funding is a weak predictor of publication output for individual researchers, but a strong predictor at the institutional level, where regional differences further influence how effectively funding translates into high-impact computer science publications.

👉 Data processing & cleaning
👉 Final analysis report

---

Dolendar System Requirements

Scope: Translated a 2-hour client consultation into a comprehensive, structured requirements document.
Deliverable: Clear user stories, functional/non-functional specs, constraints, and acceptance criteria to guide development.

👉 Full report