Development of AI algorithms for the analysis and interpretation of histopathology whole-slide images with applications in cancer diagnosis, grading, biomarker development, treatment response prediction, metastasis prediction, detection of gene mutations, tumor ROI selection, and others.Developed solutions for segmentation and classification of whole-slide images using modern AI architectures such as U-Net, Mask RCNN, ResNet, EfficientNet, Multiple Instance Learning (MIL), and transformer-based Foundation Models (DINOv2, Gigapath, UNI, CanvOI) Coordinated/managed the development of cloud-based software workflows for AI and imaging research at scale. Managed a team of researchers, software engineers and interns.

Development of AI models for automatic image/video analysis with applications in large-scale deployments of security cameras. Developed Convolutional Neural Networks such as Inception, VGG, and ResNet for image classification and YOLO for object detection. Also used Neural Architecture Search (Google's AutoML) to optimize performance for certain applications. The projects required extensive use of Python, PyTorch, TensorFlow, Keras, and Google's AI platform. The projects culminated in the successful deployment of AI models for commercial use. The models currently operate 24/7 processing millions of video frames daily. Projects required extensive cross-functional work with Software Engineering and Dev Ops teams to deploy models using Google Cloud Platform, Docker, Kubernetes, and scalable clusters of Nvidia GPUs. Managed a small team of engineers and data scientists who designed and developed AI-based applications to enhance the services offered by the company

Developed machine learning algorithms for Points-of-Interest recommendation with results displayed dynamically on a map while driving. Used a combination of Collaborative Filtering (matrix factorization) and Content-based methods. Also contributed to ongoing work to predict driving routes and destinations using user's historic data and machine learning algorithms. Software development in Python, Scala, and use of Spark MLlib.

Developed software for color transformations of very large collections of images derived from master copies of uncompressed movies (a single movie can have about 170,000 high-quality images or 4TB of image data). Developed image processing algorithms, database integration, data visualization tools, and parallelization for performance improvement when processing very large datasets using multi-CPU architectures

(1) Developed Machine Learning algorithms for a semantic classification of images using their descriptive text. Used algorithms such as K-means clustering, dimensionality reduction using PCA, word representations using TF-IDF, etc. (2) Developed Machine Learning algorithms for a semantic classification of tweets. Used algorithms such as Explicit Semantic Analysis (ESA) and an innovative Extended Bayesian approach. (3) Developed algorithms that use declarative first-order logic to predict user intent in media-sharing scenarios (connections between phones, tablets, TVs). Developed a Datalog-based inference engine and a deductive database. The project was implemented as a prototype for demonstrations in Android phones.(4) Managed a small team of researchers, software engineers, and research interns

Developed architectures and protocols to integrate user devices (computers, phones, TVs, game consoles, etc.) as part of the international UPnP and DLNA standardization projects. Worked as Microsoft's lead architect/contributor working alongside teams from companies such as Sony, Samsung, Panasonic, Intel, HP, and many others. Developed prototypes as proof-of-concept. Developed test solutions. Managed some of the engineering teams that successfully implemented these standards in Windows 7 and 8. Designed an analytics platform to generate insights on audio/video streaming usage patterns (in the context of home networks). Managed the team working on data extraction, transformation, and loading. Developed data visualization software and generated reports. The platform was successfully deployed in Windows 8.Several others projects in areas like video streaming, standardization of video/image metadata, and the development of HTML-based interactive TV applications.

Develop software prototypes for interactive Digital TV applications using frameworks based on HTML/JavaScript and alternatives based on Java. Participate as technical contributor to standardization efforts related to Home Networking and Digital/Interactive TV scenarios.

Research and develop new data transport methods and protocols for digital TV channels. Developed solutions for encoding and transferring of: electronic program guides (EPG), HTML/JavaScript content, large image databases, etc. The EPG protocols were contributed to ATSC and became a US standard.