At IBM, I build production machine learning pipelines for Instana and Watson AIOps, turning research models into scalable log anomaly detection systems. I increased log-processing throughput from 500K to 10M lines per training run while reducing model memory by approximately 88%.
I work closely with IBM Research to bring experimental models from Jupyter notebooks into production, including MLflow deployment, evaluation design, multi-tenancy, and migration from Elasticsearch to ClickHouse. I also detected data drift with over 90% accuracy using Bayesian inference.
Previously, I built healthcare prediction models at the Keck School of Medicine and backend systems at Avaya. My work spans Python, Spark, Kubernetes, Docker, cloud-native deployment, data infrastructure, and applied machine learning.
