John Passarelli Email & Phone Number

Sydney, Nsw, Au

Northville, Michigan, Us

New York, New York, Us

Joined as the company's first full-time hire, playing a pivotal role in establishing the machine learning processes. Responsible for all model development from inception to deployment. See AI solutions at this link: https://foxyai.com/foxy-ai-model-library/

At a high level:I developed and implemented algorithms (typically deep learning) to aid inspection of well pads, oil refineries and power lines.I tweaked, adapted, and implemented a variety of state of the art algorithms from computer vision literature for product solutions and when necessary develop my own algorithms. Link to patent: https://patents.justia.com/patent/11468538Avitas was originally a GE Venture and was later acquired by Baker HughesMore Detailed Level:I implemented a variety of SOTA object detection algorithms including YOLO, Mask R-CNN, and EfficientDet to detect objects in RGB and thermal images for various domains (i.e oil tanks, power lines, X-Ray plane parts, etc.)In order to detect methane plumes in thermal video, I developed a 3D convolutional segmentation neural network that took both RGB images and optical flow as input. The 3D convolutional segmentation neural network was also used to detect corrosion in thermal videos of pipes. The pipes were expensive to unsheath and thus the algorithm would detect corrosion based on the temperature changes of the pipe as fluid flowed through the pipe.I implemented a U-NET inspired segmentation algorithm to detect corrosion in RGB images. This algorithm was also used to segment liquid levels in thermal images of oil tanks and RGB images of chemical tanks to gauge how full the tanks were.Above mentioned algorithms and others were used to detect variety of damages including oil leaks, open electrical panels, debris, open thief hatches, holes in utility pole, etc.In all these situations I proposed an algorithmic plan, managed the data, trained the models, tested and evaluated models, and aided with pushing models to production.

Lexington, Ma, Us

Social Media Natural Language Processing Analytic​ ● Interactive DJango/Python​ web tool for analyzing a social media corpus● Gender prediction: convolutional neural network with keras/theano ​using word embeddings● Text classification: uses sklearn​ naive bayes to classify text as opinion/information etc.● Other algorithms include: similar image search, search suggestions, similar user search, mostindicative content, follow back, basic statistics/filteringInexact Graph Matching​● Best effort graph matching integrated into a system using AWS, Elasticsearch, PySpark, andPostgreSQL● Publication: https://aisel.aisnet.org/hicss-51/da/decision_support_for_complex_networks/2/MIT Beaver Works Summer Institute● Helped design and teach summer courses● Material included: language processing with recurrent neural networks, image processingwith convolutional neural networks, support vector machines, decision trees

Lexington, Ma, Us

• Created a Web-Based Big Data Analytic: o hosted on mini cluster o Input: CSV and JSON files o Dynamically manipulates data using common statistical algorithms o Displays data as table, map, or statistically relevant numbers • Uses PySpark cluster computing framework and Hadoop Distributed File System• Developed using Python/JavaScript/AJAX/JSON/HTML/CSS/KML/Google Earth

Armonk, New York, Ny, Us

Develop, debug and test client side applications using JavaScript/Dojo/Dijit/HTML/XML/JSON.

Winston-Salem, Nc, Us

Develop genetic algorithms in Matlab to simulate brain networks. Analyze the networks created by the genetic algorithms.

Wells, Maine, Us