Vishwanath A Sardeshpande

Designed and deployed Vertex AI Pipelines on GCP to automate ML workflows, including GenAI-based image annotation for dataset creation. Managed and debugged legacy Airflow DAGs during migration to Vertex AI, improving reliability and operational efficiency. Built APMs and monitoring dashboards using Datadog and Looker Studio to track pipeline health, latency metrics, job coverage, and dataset generation from BigQuery, enabling faster insights and incident response.

Conducted a comparative study of GNNs (GAT, GCN), benchmarking on synthetic NetworkX datasets to identify optimal graph generation architectures. Designed and trained custom PyTorch-based GNNs, including generative models that achieved 95% accuracy in producing structures like spanning trees, outperforming baseline GCNs. Developed custom loss and reward functions to improve convergence and predictive performance on large-scale graph data.