Notes on data engineering, AI and building things end to end.
Dropping every row with a missing value feels safe — it’s often the worst option. Here’s the common mistakes in handling missing data, what actually works, and how automated tests catch it early.
Read more →
Synthetic data is clean and unlimited — which is exactly why it hides the mess real data always contains. Here’s what training on real datasets like MovieLens and LIAR actually teaches.
Read more →
Star schema vs snowflake schema comes down to one tradeoff: normalization vs. query simplicity. Here’s why star schema wins for most analytics use cases, with real project examples.
Read more →
Streamlit vs Gradio isn’t about which is better — it’s about which matches what you’re building. Here’s how to choose, with real examples of each used for the job it actually fits.
Read more →
Ten projects in, a few patterns kept repeating: honest metrics mattered more than model choice, cheap data-quality tests were the highest-leverage addition, and a UI changed who actually reviewed the work.
Read more →
“It works on my machine” is the most common way a portfolio project fails to impress a reviewer. Here’s why you should containerize your machine learning model, even for a small project — with real examples.
Read more →
The same model predictions can look excellent on an ROC curve and mediocre on a precision-recall curve. Here’s why that happens, and which one to trust — especially on imbalanced data — with real examples.
Read more →
A model that performed well at launch can quietly degrade every month afterward. Here’s what model drift actually is, how to detect it, and when retraining is (and isn’t) the right fix — with real examples.
Read more →
A model that scores 98% in testing and fails in production almost always has one cause: data leakage. Here’s how it happens, how to catch it, and how to sanity-check your results — with real project examples.
Read more →
An accuracy number with nothing to compare it to is a claim, not evidence. Here’s why a baseline model matters more than the headline metric, and how to pick the right one — with real project examples.
Read more →