Problem
Retention teams need fast, testable signals for which customers are likely to churn, but raw event data is noisy and difficult to operationalize.
Case Study
ChurnLab
Flagship data + backend system for churn prediction experimentation
PythonFastAPIPandasscikit-learnPostgreSQLDockerAWS
Solution
Built a backend-driven analysis platform that ingests customer-behavior data, engineers features, and exposes prediction and reporting endpoints for downstream tools.
What I Did
Designed API contracts, implemented feature-generation workflows, and integrated model output into a usable service layer for product-facing decisions.
Challenges
Balanced model iteration speed with data quality guarantees and built repeatable validation checks to avoid silent training regressions.