Optimize container cold boot times for Flyte ML and data workflows.

Researched and experimented with strategies to improve ML model cold boot times on GPUs for Replicate.com's hosted inference service.

Built a serverless cloud functions platform which allows users to extend their products with custom code using drag-and-drop UI elements or a SDK. Part of Y Combinator's Winter '23 batch

Joined as the first engineer to work on a bit of everything! Built Measure Labs' frontend, backend, mobile apps (iOS and Android) and mobile SDK (iOS), internal tooling, and dashboards. Backend: AWS (Flask service on Fargate, AWS Lambda, RDS); blood pressure clinical data collection mobile app and blood pressure measurement mobile SDK: React Native, Kotlin Multiplatform, Swift.

Engineering lead for the launch of Contact Lens for Amazon Connect providing real-time conversational analytics and quality management powered by machine learning.

* Worked to launch two natural language processing services, AWS Comprehend and Translate* Crawled and scraped data for annotation to be used for training deep learning NLP models.* Implemented Redshift-based data mart to securely store and vend critical customer data to engineers and scientists and automated workflows for annotation and model training.* Got Translate certified for compliance protocols – GDPR, HIPAA, and FedRAMP. Implemented vulnerability scanning for fleets, automated host patching, VPC peering, and encryption for all data at rest and in transit. Designed workflow system to purge customer data for all terminated AWS accounts and customers who explicitly opt out of having data stored. Designed and implemented solution for managing internal DNS delegation.

* Worked on AWS Commerce Platform - Insights team which generates billing artifacts and insights for AWS services. Built system for tracking the delivery status of AWS billing artifacts using DynamoDB and SQS. Enabled us to meet customer SLAs by notifying our service teams of delivery issues proactively. Optimized DynamoDB queries so that wecan calculate the “staleness” of all AWS billing artifacts in less than a minute.* Built tool to send emails to allow email template owners to easily validate their changes in all stages of the pipeline.

* Worked on Forecasting team in Supply Chain Optimization Technologies to create framework for predicting customer demand for products sold by Amazon.* Collaborated with machine learning scientists to develop tools for running training and prediction algorithms at scale using Hadoop clusters running on EC2.* Helped improve forecast accuracy by adding new features to a sparse random forest model for predicting demand for new products. Deployed the model to regions world-wide.* Developed a Hive framework for generating metrics. Improved performance of later iterations by optimizing queries, tuning Hadoop parameters, and substituting inefficient queries with Python streaming Map/Reduce scripts.* Built Coral service and web portal for data analysts to review, approve, and release forecasts.

* Developed a Java application to wirelessly control a prosthetic hand using muscle and brain activity signals.* Decreased signal acquisition time by 15% by re-writing a protocol to eliminate ACK/NACK, reducing initial latencies.* Led a five-student team to gather requirements, implement algorithms, and conduct beta tests with amputees.