- Implemented the use of MPI jobs for large fine tuning jobs of LLMs of 100B+ parameters on Kubeflow reducing training time linearly, and developed training guides for the science team- Managed the release lifecycle for the MLOps platform for both the underlying infrastructure, and the applications through AWS CDK- Developed automation and services for onboarding teams and projects ensuring ease of the platform adoption- Developed a project template for the platform that encompassed all best practices for MLOps, reducing the amount of time it takes to onboard, train and deploy models in the platform- Designed an MLOps platform for Customer Service Science teams integrating Kubeflow, Sagemaker, and MLFlow helping standardize the model development process across the organization

- Worked on data intensive applications, IoT, and ML problems using AWS services and Open Source technologies for the U.S. Federal Government and Department of Defense- Worked directly with customers to develop data models for manufacturing quality control data, helping them gather insights, detect anomalies and improve preemptive actions- Architect and developed high throughput data pipelines using event driven architectures on the AWS ecosystem like Glue, Lambda, SQS, and Redshift, for IoT and batch data

- Enabled Kubeflow as a Continuous Learning Framework for AIDA (AI Driven Annotations) utilizing Argo Events and KNative Events with the capability of processing 100,000+ PDF documents daily per model. - Led Kubeflow Pipelines and KFServing integration with the AI-Platform in PwC Labs to drive containerization and scalability for machine learning models across the firm- Developed an SDK to abstract the usage of data layers (ADLSGen2), and metadata layers (MLMD)across the life-cycle of machine learning projects- Led Airflow integrations with the AI-Platform for projects to drive batch inference, and machinelearning orchestration- Helped with Spark on Kubernetes integration in the AI platform to replace ephemeral HDPclusters, reduced costs, and increase containerization of big data pipelines

- Created dockerized web scraping pipeline to extract tax related information from government websites, utilizing AWS Services, Scrapy and deployed in Kubernetes.-Utilized open-source libraries to create semi-supervised hierarchical topic models to auto classify government and tax documents - Replicated Relationship Extraction research papers using Tensorflow Estimators to understand their application in tax research areas; papers include LSTM networks with Self-Attention, and BERT- Provided data wrangling scripts to content testers to translate excel documents into transactional data utilizing Pandas dataframes

- Created test frameworks in Python using Selenium for Vertex Enterprise- Setup Continuous Integration using Jenkins to run Regression Tests daily- Created Asynchronous Test Framework using Celery for Performance Testing of Vertex Data Prep- Performed code reviews with peers as part of library maintenance and improvement- Setup Selenium GRID on Docker containers to improve resource utilization during regression testing - Created REST Automation components- Researched the use of Kubernetes to improve testing services deployments and maintenance

• Successfully completed several monthly reduced data captures of the regional offices for which a subset of databases and files were copied for testing purposes, while ensuring that non-public personal information was scrubbed using IBM Optim • Coded over 100+ unique JCL jobs which captured specific DB2 and IMS databases, tables or fields within those tables. The data captured in each request was used for testing in different types of projects • Collaborated with business partners to ensure on-demand data captures met project requirements, while successfully meeting tight capture deadlines• Tracked complex incidents related to data capture within the mainframe environment utilizing HP Service Manager, and solved them by finding root causes within a 24 hour period to avoid impacts in testing

Worked with a team across different locations to support 14,000 students in their dormitories with computer problems and networking problems

Worked on Performance Matrix for Aero Product Life Cycle Management Enovia 2013X Upgrade Utilized HP’s Quality Center 10.0 to quantify processes improvements in a PLM software upgrade which resulted in 2.1 million dollars in annual productivityPrioritized defects using agile methodology for Quick Hit Project utilizing HP’s quality center