In my current role, I have led the development and optimization of a Kubeflow pipeline on Vertex AI, achieving a runtime reduction from 9 hours to just 3 hours, leading to an annual cost saving of $10 million. I architected the MLOps workflow and integrated it with Google Kubernetes Engine (GKE) for scalable model training and inference. Additionally, I implemented CI/CD pipelines using Google Cloud Build and automated the pipeline with Google Cloud Scheduler, ensuring a seamless deployment process.I also spearheaded a Generative AI project, building APIs using FastAPI and Flask for model interaction. Leveraging Google Cloud Run for containerized deployment, I secured these APIs using Google Cloud Armor, Identity-Aware Proxy (IAP), and set up HTTP(S) load balancers for efficient traffic management. My role involved deploying a Weaviate vector database on GKE, along with automating secure data ingestion using Vertex AI Pipelines and Flask APIs. For this use case, I built a Streamlit application for interactive demos and handled extensive cloud networking, including VPC, subnets, and firewall policy configurations.I also focused on ensuring high-quality delivery by implementing CI/CD automation using Google Cloud Build, alongside writing regression and end-to-end tests for the Generative AI application. My comprehensive involvement in cloud networking and security, coupled with experience in GCP services like Vertex AI, GKE, and Cloud Armor, has been pivotal in delivering scalable, secure, and optimized AI solutions.I have also developed cloud log sink to route logs into the bigquery. Automated many bigquery tasks as well.

• Deployed and maintaining several Kubeflow pipeline based batch scoring models on Iguazio platform(Kubernetes based Mlops platform).• Built fully automated retrainable pipeline with using custom csv based model registry.• Implemented CI/CD pipeline from scratch with best software engineering principles including test cases andlinting checks.• Creating custom docker images for legacy ML models for placing into Kubeflow pipelines.• Implemented distributed hyperparameter tuning utility for data scientists and for automated trained pipelines withdask cluster.• Model and pipeline health monitoring

Aircraft Performance Analyst (Gas Turbine Engine Performance Analyst) (DPHM Analytics)Customer: Pratt & Whitney Canada

Aircraft Performance Analyst (Gas Turbine Engine Performance Analyst) (DATA Analytics)