Columbia University

Data Scientist/ML Engineer: LLM-Powered Mental Health Support Tool, Columbia University Practicum

January 2025 - May 2025

Designed an end-to-end workflow for a HIPAA-compliant mental health chatbot, incorporating informed consent, CBT-based interventions, and crisis escalation protocols aligned with clinical best practices. Contributed to building a prototype using Streamlit.

Established 10 safety and relevance metrics for AI mental health systems, including response appropriateness, risk detection accuracy, therapeutic alliance measures, and clinical datasets for model training.

Learning Outcomes: Python, Data Analysis, Machine Learning, AI Safety Evaluation

Research Assistant, Political Science Department, Columbia University

Project: Studied the influence of bot utilization by politicians on social media, and its effects on financial contributions and political outcomes.

Role: Web scraping for politicians and their contributions using an API; data analysis.

Learning Outcomes: Web Scraping, Python, Data Analysis, SQL

Ethics & Usage Lead at GradeMate, Columbia University

Project: GradeMate is an AI-integrated approach to support instructors in evaluating student assignments. The app leverages Large Language Models (LLMs) to read through assignments and provide feedback to students.

Role: Studied ethical considerations and developed guardrails for our model. My role involved literature review, LLM performance analysis, statistical analysis, and a systematic internal study to evaluate bias (using an API to study bias in grading based on race, ethnicity, gender, and nationality). I also outlined a UX research study for when the app is safe and ready for testing.

Learning Outcomes: Statistical Analysis of LLM Performance, AI Ethics Research, Research Design, UX Research.