Markets
International trade, macro context, alternative data, and country or industry screening.
Portfolio · Sangmin Lee
I test the assumptions behind data and turn it into metrics a decision can rest on.
With a background in international trade and business, I analyse public data, customer data, and alternative economic indicators. I care less about producing a good number than about checking how far that number can speak.
The three featured projects are the Busan urban-rail dwell analysis, telecom churn XAI, and satellite night-light GDP analysis. Work that explored an unfamiliar domain with AI help, like the Arctic route study, is set apart in Explore with less weight.
Working method
- 01Define the problem
- 02Check the data's assumptions
- 03Design the metric / model
- 04Validate and revise
- 05Scope the reading
How I work
I try to make the numbers useful before I make them impressive.
I start by asking why the question matters, then separate what I checked from what still needs another pass.
Finance
Financial statements, valuation logic, sensitivity checks, and assumption audits.
Analytics
Python, statistics, regression, classification, explainable AI, and survey analysis.
Execution
Dashboards, reports, MVP flows, operating checks, and clear next-step recommendations.
Selected evidence highlights
The featured projects' numbers come first.
Only the three cases a recruiter should see first are kept here. Work that is harder to defend in depth in an interview, like the Arctic route study, sits lower in the Explore section below.
Evidence
0.96 h dwell at Haeundae
What I worked on · Constructing a variable the source data does not contain, finding the first metric broken, and repairing it before reporting.
Caveat · In-progress work ahead of a September 2026 submission; dwell is estimated from gate records, not measured mobility data.
Evidence
7,043 telecom customer records
What I worked on · Customer segmentation, explainable classification, and action-oriented retention planning.
Caveat · The 5.0 percentage-point churn reduction is a proposed target, not an achieved result.
Evidence
R-squared = 0.819
What I worked on · Testing night lights as an auxiliary economic indicator, while separating data QA from model claims.
Caveat · This is explanatory fit in a simple model, not prediction accuracy or causal proof.
Selected work
Three featured projects come first.
This first screen leads with analysis I can defend myself. Each case shows the question, how the data was checked, and how far the conclusion reaches.
Explore / Archive
Unfamiliar domains and side projects carry less weight.
The Arctic route study was extended with AI help as exploratory work, so it is left out of the featured set. It stays here as supporting material that still shows habits of reproducibility, stated limits, and building systems.
Continue the conversation
I want to work where numbers have to become clear next steps.
Open to internship and entry-level roles near data analysis, financial and market research, fintech, strategy, BizOps, and service planning.