Mickey Chalupa

AI/ML Engineer with 5+ years of experience developing custom machine learning and computer vision solutions across industrial production lines, startups, research environments. Skilled in building and deploying models end-to-end—cloud to edge—with a focus on real-time tracking, facial recognition, and traditional computer vision. Adept at working with multi-modal data across disciplines to rapidly prototype, optimize, and scale ML systems from scratch.

Experienced in support, deployment, and maintenance.

Self-employed - Self-employed - Atlanta, GA

(2024-12)

• Developing and integrating a deep learning industrial computer vision for tile defect detection, cataloging, and sorting to an existing production line.
• Developed a personal UV monitor to track daily UVA/UVB exposure by combining IMU orientation data and intensity
• Integrated a haptic feedback system to a handheld virtual pinball machine including solenoid, rumble motors, and ESP32 MCU
• Developing an FDA drug approval predictive model based on chemical composition and structure.

AI Engineer - Ai.Fish - Atlanta, GA

(2024-02 - 2024-11)

• Python, Pytorch, Git, OpenCV, Azure DevOps, scikit, data and video pipeline integration, model CI/CD
• Developed and optimized deep learning Pytorch-based video-tracking models for variable lighting and complex scenes, improving real-world tracking accuracy by 25%.
• Continuously refined models and datasets by detecting tracking failures, annotating the failed cases, and integrating them in future models, while maintaining performance on base cases.
• Deployed models across hybrid Azure cloud and Nvidia Jetson edge, optimizing the model for real-time inference with TensorRT, pruning, and ingestion optimization; built reproducibility framework and performance metrics to accelerate iteration cycles.

AI Developer - MOAC - Atlanta, GA

(2022-02 - 2024-01)

• Python, TensorFlow / Keras, Git, OpenCV, Azure DevOps, scikit, data and video pipeline integration, model CI/CD
• Rebuilt Dense Landmark deep learning facial-tracking system with custom ResNet-50 model, boosting stability and accuracy for facial analysis.
• Architected RNN-based temporal model to cut frame-to-frame landmark jitter by 80%, enhancing video coherence, using a custom PyTorch based DDP training loop.
• Led synthetic dataset pipeline via Blender scripting, expanding lighting and persona diversity for stronger model generalization – ranging from 100,000 to 800,000 images.
• Developed a distributed data pipeline, including sequencing and data augmentation, to cut down model training time from a 7 days to 12 hours.

Graduate Research Assistant - Kennesaw State University - GA

(2021-03 - 2022-01)

• Python, PyTorch, TensorFlow / Keras, scikit-learn, Git
• Developed dynamic client-server matching algorithm for federated learning on mobile devices; submitted to ICNC 2024.
• Built GAN-based synthetic medical image generator to evaluate privacy risks (WACV 2021 extension).
• Integrated VR-driven simulator for autonomous driving attention study; supported data collection for SSRN 2022 publication.

Computer Vision Engineer - Marietta Non-Destructive Testing - GA

(2017-04 - 2020-11)

• C++, C#, OpenCV and SQL
• Developed SCADA software, along with PLC integration and control, and developing driver interfaces.
• Engineered CV algorithms using for x-ray and ultrasound defect detection; integrated new imaging hardware into legacy inspection systems.
• Optimized existing systems to meet evolving client requirements, improving defect-detection accuracy by 30%.
• Conducted hands-on training and provided technical guidance to stakeholders and end users.

M.S. - Computer Science – AI/ML - Kennesaw State University (2022-05)

B.S. - Electrical Engineering - Georgia Tech (2012-12)