At BSH Hausgeräte GmbH (Bosch Group), I built a Python Data Quality Notification System monitoring 40+ Airflow DAGs and 30+ SAP Datasphere task chains. It reduced pipeline retry rates by 66% and achieved 91–94% anomaly-detection accuracy.
At Ignitus, I deployed 9 FastAPI model services on Kubernetes at about 120 ms p95 latency. I also maintained MLflow tracking and model registry, and automated GCP training pipelines that cut model retraining from 5 days to 14 hours.
At Grip Mechatronics Pvt Ltd, I built C++ and Python computer-vision inspection algorithms for production lines, achieving 96.4% precision at 45 FPS. Calibration and hard-negative mining reduced false rejects from 7.2% to 1.8%.
In my projects, I built a medallion lakehouse on S3, Delta Lake, and Spark that cut training-data preparation from one week to 27 hours; the MVP sold for $13,000 via Contra. I also built evaluation tools for agent overfitting and an LLM data analyst with sandboxed pandas execution and a deterministic grounding gate.

