The tutorial develops an end-to-end sentiment analysis workflow using the Stanford NLP IMDb Large Movie Review Dataset, comparing classical machine learning with parameter-efficient transformer fine-tuning. It establishes a reproducible environment and audits the dataset for class ordering, review-length skew, duplicate leakage, and preprocessing artifacts before training a TF-IDF and Logistic Regression baseline. The workflow then fine-tunes DistilBERT with LoRA through PEFT, evaluating it using accuracy, macro-F1, ROC-AUC, confusion matrices, and ROC curves, and examines threshold selection and probability calibration through Expected Calibration Error and reliability analysis. Beyond headline metrics, it investigates confident errors, performance across review lengths, word-level occlusion saliency, and head-versus-tail truncation to understand model predictions and long-context limitations. Finally, it uses the unlabeled IMDb split for confidence-based pseudo-labeling, compares the resulting semi-supervised model against the baseline, and saves the merged transformer for reusable sentiment inference.