ML Engineer shipping production deep learning models on GCP

Reduced inference compute cost 4x while serving 10,000 concurrent quote requests by optimizing CatBoost and XGBoost models with TensorRT on GCP. Accelerated model retraining latency to under 24 hours by automating drift-triggered pipelines with GitHub Actions and GKE shadow deployments. Cut manual claims verification time from 14 days to under 48 hours by deploying a multimodal RAG system linking accident photos to audio transcripts. Validated pricing algorithms with statistical significance by designing an A/B/n experimentation framework.

Scaled inference to process 5B+ transaction rows daily for 15M+ SKUs by parallelizing CatBoost and LightGBM on distributed PySpark clusters. Reduced model deployment duration by 90% by migrating 40+ legacy ML jobs to GCP Dataproc with CI/CD pipelines. Achieved 99.9% scheduling reliability for the global staffing forecast engine by orchestrating retraining DAGs in Airflow and adding data-quality gates. Cut time-to-insight from 5–7 days to under 4 hours by consolidating telemetry into governed Tableau dashboards.

Cut project status reporting turnaround by 40% by replacing Excel workflows with unified SQL data models and Power BI dashboards. Reduced material procurement risk by 15% by deploying regression models that predict cost overruns from historical data. Maintained zero data-loss incidents over 18+ months by building fault-tolerant ETL pipelines on GCP Cloud Functions with Airflow scheduling. Surfaced 7–10% pricing inefficiencies by segmenting 10,000+ suppliers into reliability profiles using K-Means clustering with SMOTE.