Designed and deployed a scalable end-to-end computer vision pipeline on Amazon Web Services using AWS Step Functions, AWS Lambda, Amazon SageMaker Batch Transform, and Amazon DynamoDB for production-scale sneaker analysis. Developed a multi-stage ML/CV pipeline with person detection, custom PyTorch sneaker segmentation, and FAISS embedding-based sneaker brand/model classification. Built modular serverless inference services with one Lambda per model to improve scalability, maintainability, and parallel processing. Optimized the pipeline for high-throughput processing of 10k+ images, reducing inference time and infrastructure costs in production.

Designed and deployed real-time computer vision pipelines for basketball analytics using AWS and GCP. Developed 2D and temporal models for detection and pose estimation on live video streams. Reduced inference latency by using ONNX optimization and GPU acceleration. Built scalable infrastructure with automatic vertical scaling for handling live streaming workloads. Built and deployed multimodal systems using Vision-Language Models (VLMs) for OCR and visual understanding tasks.

Led Smart City AI projects including face recognition, license plate detection, and smart parking. Built real-time pipelines using Docker and GPU-based inference systems. Designed scalable architectures for on-premise deployments.

Developing Computer Vision projects using MLOps tools in a Big Data team. Led development of large-scale computer vision systems in a Big Data environment. Trained and deployed models using NVIDIA TAO and DeepStream for high-performance inference. Converted models (Keras/YOLO ONNX TensorRT) for production deployment. Managed multi-GPU training pipelines for medium and large-scale datasets.

Leading and developing projects of Machine Learning and Computer Vision for Education, Retail, Ergonomics 4.0 using open source Deep Learning frameworks from collecting data to deployment with API. In other one for a PUCPR university we implemented a face expression, gaze estimation and people recognition system in order to measure attention and students’ emotions in a classroom.