Selected Applied Work
Selected projects with the questions, methods, and artifacts behind my work in model evaluation, data analysis, and usable computational systems.
Data and Models
Prepared COMI-LINGUA for entity-level evaluation, fine-tuned mBERT and XLM-RoBERTa, and compared them with zero-shot GPT-4o and Claude 3.5 Sonnet under a common protocol. The repository includes data preparation, evaluation, and error-slice analysis code.
Built a benchmarking pipeline for more than 15 regression models using over 3,000 job postings. Experiments were tracked with MLflow and compared using MAE, RMSE, and R². SHAP analysis was used to interpret associations between model predictions and features including geography, firm size, and skill requirements.
Modeled statistical associations between socioeconomic indicators and violent crime rates across 1,994 communities. Cross-validation yielded R² = 0.85 for the selected predictive model. The analysis does not make causal claims, and the results were presented through an interactive Streamlit interface.
Applied Systems
Contributed to a team-built platform for campus resource allocation and scheduling under shared constraints. The project covered role-based access, conflict resolution, RESTful APIs, integration testing, containerized deployment, and automated CI/CD workflows.
