/projects$cat michelin-bottleneck.html
Predictive Bottleneck Modeling
Michelin Lanka · AI/Data Engineer intern · 2024–2025
predicting bottlenecks
I studied the factory floor to understand how the line actually moves, then built an operational workflow simulation in SimPy, generated synthetic datasets from it, and trained a sequential XGBoost model to predict production bottlenecks before they happen.
automating the reporting pipeline
The reporting workflow relied on manual Excel ingestion — slow and error-prone. I replaced it with an automated FastAPI + SQLite ETL pipeline, with a hashing mechanism to prevent concurrent data duplication. Reporting time dropped from hours to seconds.