What Hardik demonstrated

Built an end-to-end turbine yield predictor using leakage-free sensor inputs, a time-based held-out split, and honest evaluation. Compared a LinearRegression baseline with a tuned RandomForest, then documented the model’s accuracy, limits, and monitoring plan in a decision-ready model card. The final artifact shows strong judgment around data leakage, validation, and operational deployment risk.

• Leakage-Free Feature Selection
• Held-Out Evaluation Discipline
• Residual-Based Error Analysis
• Plain-Language Model Communication