Run this notebook: Open in Colab Open in Kaggle

Phase 17: Debugging & Troubleshooting — Start Here¶

Diagnose and fix AI system failures systematically — from data issues to slow inference to hallucinating models.

Why This Phase Matters¶

90% of AI project failures are not model failures — they are data issues, evaluation mistakes, or infrastructure problems. This phase teaches a systematic debugging mindset.

Notebooks in This Phase¶

Notebook	Topic
`01_debugging_workflow.ipynb`	Systematic AI debugging methodology
`02_data_issues.ipynb`	Data leakage, class imbalance, drift detection
`03_performance_profiling.ipynb`	Profile slow code, CUDA bottlenecks, memory
`04_model_debugging.ipynb`	Overfitting, underfitting, gradient issues
`05_error_analysis.ipynb`	Confusion matrices, failure mode analysis

Common AI Bugs Taxonomy¶

Category	Examples
Data bugs	Train/test leakage, label noise, class imbalance
Training bugs	Wrong loss function, LR too high/low, batch size
Evaluation bugs	Wrong metric, leaky evaluation, benchmark overfitting
Inference bugs	Wrong preprocessing, tokenization mismatch
LLM-specific	Hallucination, context overflow, prompt injection

Prerequisites¶

Machine learning basics
Model evaluation (Phase 16)

Learning Path¶

01_debugging_workflow.ipynb      ← Start here
02_data_issues.ipynb
03_performance_profiling.ipynb
04_model_debugging.ipynb
05_error_analysis.ipynb