Vivin Dsouza

Engineered an end-to-end elevator predictive maintenance pipeline evaluating multi-class failure modes using raw accelerometer time-series and wide KPI metrics.

Optimized

LightGBM and XGBoost classifiers to achieve a 0.8714 F1 score and 99.07% accuracy, utilized Optuna hyperparameter search and per-class decision threshold tuning to mitigate multi-class imbalance. Designed a trip-count Remaining Useful Life (RUL) estimation framework coupled with an algorithmic Health Index mapping to derive real-time equipment risk status lookups.

Machine Learning Research Intern at Schindler, Switzerland (2026-02 – Present)

• Engineered an end-to-end elevator predictive maintenance pipeline evaluating multi-class failure modes using raw accelerometer time-series and wide KPI metrics.
• Optimized LightGBM and XGBoost classifiers to achieve a 0.8714 F1 score and 99.07% accuracy, utilized Optuna hyperparameter search and per-class decision threshold tuning to mitigate multi-class imbalance.
• Designed a trip-count Remaining Useful Life (RUL) estimation framework coupled with an algorithmic Health Index mapping to derive real-time equipment risk status lookups.

Major in Computer Science and Engineering in Computer Science and Engineering – National Institute of Technology Karnataka, Surathkal (2024-08 – 2028-05)