At Way2 Technology, I developed an NLP-based fraud detection model for electricity consumption irregularities that achieved a 98% F1-score. I also optimized a MobileNetV2 model for on-device building facade classification, reaching 93% accuracy while reducing its size by 40.8%.
I deployed an ML model on Amazon SageMaker Serverless Inference and designed monitoring for data drift and model quality. Previously, at Colibri Interfaces and Technology, I led development of an eye-tracking system for digital accessibility and worked on assistive-device firmware, an Android application, and PCBs.
