Joyce

PORTFOLIO

Back to Projects
Live2026

TikTok A/B Testing Suite

View CodeLive Demo

ABOUT THIS PROJECT

Simulates a real-world A/B experimentation suite across 4 concurrent TikTok product tests — with rigorous statistical analysis and an executive-grade Streamlit dashboard.

Covers Welch's T-Test for continuous metrics, Chi-Square for conversion rates, Cohen's d for effect sizes, and power analysis for sample sizing. The Streamlit dashboard is designed to read like an internal TikTok product review doc — black header, gold accents, Playfair Display, and green/red verdict cards for instant decision clarity.

WHAT I LEARNED

Welch's T-Test over Student's T-Test is almost always the right default — assuming equal variance between control and treatment groups is rarely justified in real product experiments

Cohen's d made effect sizes legible to non-statisticians — a p-value alone doesn't tell you if a result is worth shipping, but 'medium effect' does

Running 4 concurrent tests exposed the multiple comparisons problem — Bonferroni correction feels conservative but it's the honest approach when testing several hypotheses at once

Dashboard design for stakeholders is a different skill from analysis — the verdict cards weren't aesthetic choices, they were communication choices

Power analysis before running a test is the step most people skip — it's the only way to know if your sample size is actually large enough to detect the effect you care about

TECH STACK

PythonStreamlitSciPyPandasNumPyPlotly

LIVE PREVIEW

Having trouble viewing? Open in new tab

Back to all projects