Hand2Num

Why I built it

Handwritten digits offered a focused way to learn the full path from image cleanup to model output.

Problem being solved

Raw handwriting varies in scale, position, contrast, and noise before a classifier can produce dependable numeric output.

Key challenges

• Normalizing inconsistent handwritten input.
• Keeping preprocessing aligned with model expectations.
• Inspecting failure cases beyond aggregate accuracy.

Design decisions

• Make preprocessing stages independently inspectable.
• Keep inference output structured for downstream use.
• Use visual examples to reason about misclassification.