° Identify ML opportunities to solve business problems and optimize workflows with stakeholders.° Support production ML system and retrain XGBoost and BERT models as needed.° Introduced NLP model development at Porch, leveraging Label Studio to streamline ground truth gathering.° Created NLP token classification models to extract key data from documents, which enhanced Porch Insurance’s pricing models that resulted in lower costs for customers.° Built text classification models to improve the performance of downstream data extraction processes.° Run and analyze A/B tests on Machine Learning models.° Redesigned the prediction data logging architecture to efficiently process large datasets for ML model retraining and analysis.° Spearheaded the adoption of BentoML and collaborated with its community to deploy Porch’s first Python model service, reducing model deployment time by 90%.° Adopted Feast feature store to enable feature reuse, reducing 80% of Kubernetes pod memory usage.° Proposed and deployed Union Cloud (Flyte managed service) with Union AI, empowering ML practitioners to automate their model-building pipelines.Technologies used: Python, Numpy, Pandas, Scikit-Learn, XGBoost, BERT, BentoML, Feast, Flyte, Airflow, Kafka, Google Cloud Platform, Java, Kotlin, Scala, Spark, Spark ML.

° Overseeing the development and deployment of the LINC system, https://linclion.org/, for Lion Guardians, http://lionguardians.org/.° Developed and deployed Computer Vision models and services powering LINC.° Acted as a liaison between software engineers and research scientists.Notable contribution:When I joined Lion Guardians as a volunteer Machine Learning Engineer, the LINC system had been unusable for 1.5 years despite efforts from previous developers. Key functionalities were broken, leaving the system inoperative. I collaborated with software engineers and research scientists to fully restore the system within 6 months, bringing it back online and operational.

° Collaborated with cross-functional teams in delivering products.° Collaborated with data scientists to design, build, and deploy scalable ML web services and pipelines for scoring millions of images, while contributing to model development.° Proposed and implemented a proof of concept using Stanford NER to classify named entities as brands or people, improving on rule-based system limitations.° Designed and implemented the Getty Knowledge Graph database to capture entities and their relationships, enabling smarter, more efficient search results for customers.° Developed and maintained web services using Apache Solr to power search functionality for millions of Getty documents.Technologies used: Java, Scala, Python, Numpy, Scikit-learn, Pandas, Flask, gunicorn, Spark, aiohttp, AWS.

Responsible for developing and supporting underwriting applications for large insurance company.

° Helped clients to deliver business values using Agile Principles.° Developed Web application for Gordon Food Service.

Developed enterprise application for clients.

° I served as application architect for the development of enterprise servicing and collection Web applications at CitiMortgage. ° I mentored developers in coding best practices and design patterns. Both servicing and collection applications were the first to adopt Domain-Driven and Test-Driven Design and Continuous Integration (Cruise Control). The servicing application was integrated with Tibco EMS to cross-sell Citi products to CitiMortgage customers.° My group was the first to explore and adopt agile development (XP) for both onshore and offshore. We utilized Xplanner to facilitate the development process.

I co-lead the development of administrative portal Web application for decision engine project. It was the largest project ever initiated by Principal Residential Mortgage (then acquired by Citigroup). The Administrative Portal was the primary tool for managing the decision engine knowledge base (rules).