I'm a math student at Baruch College interested in data science, machine learning, and statistical modeling. I like working with messy real-world data and figuring out what the numbers are actually telling us.
I'm particularly interested in model evaluation, subgroup differences, and the assumptions and artifacts that can get lost behind a single performance metric.
I also design and lead workshops as the Workshop Director at Baruch's Machine Learning & Data Science club, check the workshop material out here.
-
Who Returns to SYEP? Predictors of Reapplication to NYC's Summer Youth Employment Program
Built a longitudinal pipeline and cluster-robust logistic regression models on ~130K NYC DYCD participant records, finding that hours completed and first-choice placement predict whether youth return, while self-reported program benefits don't.
→ https://github.com/soniatyburczy/syep-reapplication -
Structural Predictors of Chronic Absenteeism in NYC Public Schools
Compared statistical and tree-based models on NYC DOE data to study structural drivers of absenteeism and the impact of an administrative data shift.
→ https://github.com/soniatyburczy/absenteeism-structural-predictors -
Looking Beyond Accuracy: Subgroup Recall in Adult Income Classification
Compared logistic regression and a neural network, showing how nearly identical F1 scores can hide meaningful subgroup recall differences.
→ https://github.com/soniatyburczy/adult-income-subgroup-analysis -
Fine-Tuning LLaMA-2-7B-Chat with QLoRA
Fine-tuned LLaMA-2-7B-Chat with QLoRA for structured cover-letter generation, improving grounding, semantic similarity, and repetition.
→ https://github.com/soniatyburczy/llama2-qlora-finetune