Problem
Real datasets can be sparse or constrained, making it hard to run safe, repeatable churn experiments at scale.
Case Study
Independent Study: TabDiff
Research-backed engineering for synthetic tabular data generation
PythonNumPyPandasJupyterscikit-learn
Solution
Implemented and tested a TabDiff-inspired workflow to generate synthetic tabular records and benchmarked downstream prediction behavior against baseline datasets.
What I Did
Framed the experiment design, implemented preprocessing and evaluation scripts, and documented tradeoffs between fidelity and utility.
Challenges
Controlled for data leakage risk while keeping synthetic outputs statistically meaningful enough for practical experimentation.