At NOVA, I implemented a serverless Amazon Bedrock service that extracts gas meter readings from field images and returns validated responses. By evaluating model alternatives with the client, I reduced inference response time from approximately 15 seconds to 5 seconds while balancing latency and accuracy.
I also designed and implemented an Amazon Personalize recommendation platform, including data pipelines, model training, and inference APIs. I contributed to approximately 11 Generative AI assessments and participated in 15+ AWS Well-Architected Reviews, translating findings into solution proposals and customer recommendations.
