At Amazon Alexa, I fine-tuned the Alexa+ Setup model on data from 10k smart home products, cutting legacy-flow fallbacks 68% and contributing to a 17% lift in setup success. I also co-built the output layer that converts LLM intents into API calls and delivered Post-Setup and Bulk Post-Setup flows.
At Bajaj Finserv Health, I built document-fraud detection and risk scoring models and automated 40% of OPD claim adjudication. My Georgia Tech projects include benchmarking hierarchical world models and building an adaptive vision transformer inference server; I’ve also authored or co-authored research publications in machine learning and computer vision.
