Hire Srivatsan S. - ML Engineer - Bangalore

* Played a crucial role in developing the extraction engine using multimodal models for domain-specific extraction of financial wealth and financial spreading documents. Achieved over 90% field-level accuracy in financial spreading, and around 75% field-level accuracy in financial wealth documents. * Developed multiple inference services to handle large volumes (approximately 100-150 per day) of large documents with an average size of 50-60 pages. * Optimised inference services for CPU and RAM utilization by reducing CPU usage by 40% and RAM usage by almost 70% while maintaining throughput. * Developed a data mapping tool to auto-annotate documents based on historical data, thus eliminating the need for annotation.

* Played a crucial part in developing an MLOps platform using Kubeflow, Kubernetes Apache Flink and Feast (as a feature store), able to host jupyter notebooks, infer models and post production monitoring for data and model drift. * Developed a custom Kubernetes operator for Apache Airflow for a single-node-per-pod run, to maximise cost savings and efficient scaling.