The Problem

Most teams run A/B tests without proper power analysis, leading to underpowered experiments and false positives. Black-box tools hide the statistics, making results unauditable — and offer no way to monitor experiments safely while data is still collecting.

What I Built

• Built a custom statistical engine from first principles covering pre-experiment power analysis (Cohen's h for proportions, Cohen's d for continuous metrics), hypothesis testing (z-test, t-test, Mann-Whitney U, chi-squared), and sequential monitoring with SPRT — implemented directly from statistical formulas, not wrapped libraries
• Five-page React dashboard: experiment wizard, power curve visualization across MDE values, live results with effect size and confidence intervals, sequential monitoring view, and full experiment history
• FastAPI backend with PostgreSQL persistence and Docker Compose deployment — production-ready architecture with CSV upload support for external data sources

Impact

Gives data science and product teams a transparent, audit-ready experimentation platform where every statistical decision — sample size, test choice, stopping rule — is configurable, visible, and explainable.