Software Engineer II - ML/ Data @ Microsoft (July 2022 - Present): Used a internal PaaS that abstracts Azure Serverless Function and AzureML service to preprocess and train Logistic Regression and Random Forest to classify messages in M365 Admin Center into High, Medium or Low to help Admins triage msg based on its significance to their tenants. Each functionality is encapsulated into its own module of code - read data from Az Datalake and preprocess, train LR and RF model, and upload model joblib files to Az Blob Storage. Coded inference logic using an internal managed Infrastructure called Polymer that provide support for metrics & logging, creating & monitoring deployments and hosting inference as a service endpoint. Software Engineer III - ML/Data @ Walmart (May 2021 - July 2022): ETL of highly sensitive store employees data for NDA project: Coded custom Airflow DAG & Python Operators to auth with on-prem IaaS MS SQL Server via Kerberos (krb5), configuring a krb5 client in Docker running Airflow, & initiating a krb5 ticket in runtime via python’s subprocess lib & credentials from Airflow conn. encrypted in PostgreSQL metadata db. Coded a bi-weekly sliding window delta logic and fixed Kubernetes pod OOM bug by designing task specific executor RAM & CPU, & by extracting millions of rows from MS SQL & loading into Google Big Query in parameterized chunks, consumed by PowerBI & for model training. Software Engineer II- ML/Data @ Walmart (Jan 2021 - May 2021): Time Series Sales Forecasting: Developed pyspark libraries on Databricks, to extract item level register scan sales from on-prem to cloud, joining, grouping by and aggregating to create daily sales for each store & dept. Improved job speed by 75% via configuring spark dynamic overwrites. NLP Classifier of Prescribers: Developed a scraper app of US medical boards and built CI/CD pipeline on Azure DevOps to deploy a multi-class classifier model built on BERT backbone to Azure Databricks. Coded an Azure Event Hub producer to publish model predictions as a stream. Protégé effect, - Best way to learn something is to teach it: Articles and Videos I published: Youtube Video: https://www.youtube.com/watch?v=6THcISTfvF8&t=17s (Leetcode Problem Solving) Linkedin Article: https://www.linkedin.com/pulse/reorder-data-log-files-hary-krishnan/ (Leetcode Problem Solving) (All Linkedin Articles: https://www.linkedin.com/in/harykrishnan/detail/recent-activity/posts/)