Working on Health Data Intelligence

Designed AWS-based Big Data systems for high-impact processing of a 200 petabyte-scale advertising data store, managing over 50 clusters, each with up to 60 EC2 instances (~60MM instance hours).Created a scalable and extensible Generative AI framework using Amazon Bedrock to build NLU based SQL Generation tool. This supported developer deep dive investigations across 200 PB of multi-cluster data and reduced p99 resolution times from 10 days to 5 days.Enhanced and maintained Deterministic Attribution system (annual budget $18MM), which provides advertisers with conversion reporting through attribution algorithms and data processing pipelines at an event grain with ~40 million events processed per hour.

Trained individual machine learning agents with the collected data (location, ground speed, and motion classification) to build a Just In Time Adaptive Intervention system to promote active lifestyle and reduce sedentary behaviors by alerting (‘nudging’) participants through Apple notifications.Developed an Apple Watch and iOS application data collection package for monitoring over 30 research participants for a research paper.

Implemented large scale computer vision network using over 800 street cameras across the state of Delaware for real-time traffic data logging, capturing up to 1,000 data points daily from each camera.Transformed and analyzed the collected data to inform the Department of Transportation when traffic was caused by flooding to help improve road conditions statewide.

Preprocessed dataset of ~50 thousand code and unit test pairs scraped from Github.Finetuned state-of-the-art encoder-decoder models like seq2seq and word2vec on the data to produce a unit test given a Java code block as a way of benchmarking our proposed model’s performance against traditional NLP methods.