@enghamza

I Trained a Clinical AI Model on Corrupted Hospital Data. Here's What Broke It Most.

Mar 23 · 6 min read · A BSAI student's first real project — real ICU data, real findings, real deployment I want to be honest with you before we start. Three weeks ago I did not know what MIMIC-III was. I had never loaded

Train/Val/Test Splits – 3 Experiments That Stopped Me From Lying to Myself

Mar 8 · 3 min read · I used to think splitting data was boring: just train_test_split(test_size=0.3) and done. Then I ran three tiny experiments and watched my test scores go from "realistic" to "too good to be true" beca

Underfitting vs Overfitting – 4 Mechanical Experiments That Made It Automatic For Me

Mar 8 · 4 min read · After finishing signal/noise and bias/variance, I could kinda explain underfitting vs overfitting… but I still had to think about it. I wanted it to be instant — like looking at a car and immediately

Bias vs Variance – 4 Simple Experiments That Finally Made It Click For Me

Mar 6 · 4 min read · ## The Setup (same for all four projects) We always use the same underlying truth: ```python true_function = lambda x: 1.5 * x**2 - 2 * x + 0.8 y = true_function(X) + np.random.normal(0, 0.8, size)

Why Models Hallucinate Garbage — And How 5 Simple Experiments Taught Me That

Mar 1 · 5 min read · Yo homie, if you're like me a few months ago, you probably thought: "Bigger model + more data + fancy loss function = magic truth finder." Then I did five tiny experiments with nothing but NumPy, scik