Bangalore Urban, Karnataka, India
1. Managed MLOps for around 150 workflows, including ETL pipelines, Churn Propensity, Segmentation, Dormancy Winback, and Classification models, using Apache Airflow and AWS to target the weekly marketing campaigns. Handled the entire migration of the codebase along with 500 TB of data from IBM Netezza to AWS Redshift.
2. Built document processing workflows using AWS Cloud Services to classify, extract and store key information for the digitization of documents leveraging the power of deep learning for varied clients:
• ML Lead for the project on extracting PII data from ophthalmic prescriptions. Deployed an ensemble model consisting of Detectron2, GraphTSR, and LayoutLMv3 models in the client environment, reducing the turnaround time by over 70% while maintaining high extraction accuracies. Built the entire backend re-training pipeline which reduced human effort time by 48 hours per week.
• A core member of the ML team in building a solution to classify and extract confidential information from non-disclosure agreements for a global transport and logistics company.
• Revamped and incorporated the LinearSVC and Elmo biLSTM models to assist the client in classifying the identity documents and extracting PII data from them.
• Deployed an automated workflow, consisting of U2Net Segmentation and ARShadowGAN models, for cropping a car, adding a shadow to the cropped image, and placing it on a standard background, facilitating an automotive retailer-based company in reselling used cars.
3. Worked on enhancing the internal QDox platform by integrating PICK, GraphTSR, LayoutLM, LayoutXLM and LinearSVC models with the UI. Aided in optimizing the overall costs by replacing all Provisioned endpoints on the platform with Serverless endpoints.
