Debugging

Interview

Look at this code snippet — what's wrong?

Tests whether you spot data-leakage bugs reflexively. They're the most expensive mistake in ML.

Interview

Your model has 95% training accuracy but 60% test accuracy. What do you investigate?

Classic diagnostic question. Tests whether you have a structured debugging process, not just memorized fixes.

Prompt

Debug an Error

Get a structured diagnosis of a bug or stack trace.

Prompt

Fix a Prompt That Hallucinates

Reduce fabricated facts, citations, and details in model outputs.

Prompt

Fix a Prompt That Ignores Instructions

Diagnose why a model skips constraints and rebuild the prompt for compliance.

Prompt

Fix Inconsistent Output Format

Make a prompt produce reliably structured output every time.

Prompt

Fix a Prompt Missing Context

Identify what context a prompt needs and restructure it to accept the right inputs.

Prompt

Fix a Prompt That Gets Refused

Rewrite a prompt that triggers safety refusals while keeping the legitimate intent.

Prompt

Fix a Prompt That's Too Generic

Sharpen a vague prompt that produces bland, template-like output.

Prompt

Fix a Prompt That's Too Long

Compress a bloated prompt without losing intent, constraints, or output quality.

Prompt

Fix a Prompt With Wrong Tone

Adjust a prompt so output matches the voice, formality, and register you need.