Hello, I'm
A results-oriented Machine Learning Engineer specializing in the full model development lifecycle. I leverage expertise in continual learning, pretraining, and MLOps to build efficient and scalable computer vision solutions.
Download CV View My Work
I am passionate about tackling complex challenges in computer vision. My experience includes designing automated labeling frameworks, optimizing training pipelines on large-scale datasets, and architecting end-to-end MLOps solutions. I specialize in enhancing model performance through techniques like continual learning, domain adaptation, and advanced pretraining to solve real-world problems.
Designed and deployed an automated system using object detectors and SAM, which cut labeling time by 50% (saving over $100,000) and scaled training on a dataset of over 300,000 images.
Architected and delivered an end-to-end assisted labeling tool with a PySide6 UI and SAM backend, doubling labeling throughput and replacing the legacy annotation tool across the organization.
Reduced training data needs by 30% via pretraining, cut false positives by over 90% with negative instance mining, and built robust MLOps pipelines using Docker, Jenkins, and JFrog for seamless deployment.
My Master's thesis. Developed a two-stage pipeline to break thumbnail-preserving encryption, using SwinIR and a ControlNet-guided Stable Diffusion model to achieve a +73% SSIM and +33% PSNR improvement over baseline.
Architected an MAE-inspired pre-training for a hybrid convolution-transformer architecture, improving classification accuracy on ImageNet-100 from 78.6 to 80.18 on 12M parameter model through custom masked attention blocks and masked convolutions.
Investigated and developed an ensemble of 2D Attention-ResUNet architectures to improve IoU scores. Applied advanced data augmentation to mitigate overfitting on limited medical datasets, achieving a 0.8 IoU score.
I'm always open to discussing new projects, creative ideas, or opportunities.
Feel free to reach out!