Brad Kim

Brad Kim

Senior Data Scientist at Phia · New York

I build LLM agents and ML systems that help product teams make better decisions. At Phia, I built a Claude-powered analytics agent and ran the experiments behind our growth from 300K to 2M users.

Before that, I earned my master's degree at Columbia University, focusing on data science. Earlier, I spent two years as a data analyst at MIDAS IT, working on customer analytics for global sales and marketing.

Work

2025 — now

Senior Data Scientist · Phia

  • Built Phia Data, a Claude-powered Slack analytics agent with nightly evals. It serves 20+ users and cut ad-hoc data requests by ~75%.
  • Built 30+ production dashboards and standardized cross-platform KPIs, enabling self-service analytics and executive decisions on revenue forecasts, marketing budgets and the $35M Series A.
  • Led 20+ A/B experiments and owned growth analytics from 300K to 2M users.
  • Built Airflow pipelines and canonical data models that cut BigQuery costs by ~45%.
2025

Data Science Capstone Consultant · Société Générale

  • Built an unsupervised AML anomaly detection system on 861K international wire transactions.
  • Engineered 240 features and built hybrid HBOS / PCA / Isolation Forest / ECOD ensembles that caught all 32 benchmark suspicious transactions.
2022 — 2024

Data Analyst, Global Sales & Marketing · MIDAS IT, Seoul

  • Used SQL, Python and BI to expand product adoption to 80+ universities in the U.S., UK and Canada, beating targets by 200%.
  • Standardized HubSpot CRM and built 15+ Power BI dashboards on leads, web traffic and customer journeys, contributing to a 30% increase in free-trial sign-ups.
  • Analyzed university content demand and student career needs in Python, which shaped an educational platform with 30+ tutorial videos and career resources.
2024 — 2025

M.S. Applied Analytics · Columbia University

GPA 4.0. Founding board member, Columbia University Consulting Club.

2021

B.A. Communication · Boston College

Cum Laude.

Projects

Phia Data: LLM analytics agent

A Claude-powered Slack agent that answers the team's data questions automatically, with nightly evaluations to keep answers reliable. Cut ad-hoc requests to the data team by ~75%.

LLM agentsClaudeSlackevals

AML Anomaly Detection

An unsupervised detection system with risk tiers and a Streamlit investigator dashboard for 861K wire transactions.

PythonIsolation ForestECODStreamlit

Amazon Cart Abandonment

Combined logistic regression (78.4% accuracy), K-means and sentiment analysis on 600+ users. Users with low trust in recommendations were 67% more likely to abandon.

PythonK-meansNLP

Compared five classifiers. Random Forest reached 95.4% accuracy and an AUC of 0.98.

PythonRandom ForestSVM
writeup notebook ↗

Predicted ad CTR with an XGBoost model tuned by Bayesian optimization, and identified the factors that drive engagement.

RXGBoostBayesian Opt.
writeup report ↗

Diagnosed an 18% revenue drop across five U.S. cities using Tableau and regression analysis.

TableauRegression
writeup report ↗

Skills

languages
Python, R, SQL
ml / ai
LLM agents (Claude), evals, XGBoost, Random Forest, anomaly detection, A/B testing
data
BigQuery, Snowflake, PostgreSQL, Airflow, PySpark, Hive
tools
Looker, Tableau, Power BI, AWS, Azure, Datadog, Git