Hoss is a boutique data science consulting firm delivering end-to-end machine learning solutions. Both founding members are data scientists specializing in multiple fields in the algorithmic landscape, and come with many years of experience both in freelancing and full-time roles. Recent clients / projects include: ● Retail-AI: using overhead video footage of grocery store registers to count the number of items scanned by the cashier to help detect theft (fully convolutional neural net on variable length input video) ● De-ID: fooling facial recognition algorithms with adversarial generative neural networks (GANs) ● Kite: machine learning for intelligent code completion in Python (deep neural nets) ● Roofr: using satellite images of houses to segment the exact roof shape in the image (convolutional U-Net architecture segmentation network) ● Range: identifying named entities in text snippets to find commonalities between different threads (GBM over tiny labeled dataset, leveraging external data for feature engineering) ● Ridevision: using frontal motorcycle video footage to predict the time-to-collision of an upcoming vehicle, to alert the driver to potential danger and prevent accidents (3D convolutional neural net) ● TinyInspektor: using images taken from factory manufacturing lines to automatically identify and locate physical production defects (YOLO architecture neural net) ● Orbograph: examining scans of checks to determine presence of various properties (is the check 'for deposit only', etc.) ● SecuredTouch: identify when a smart-phone user is fraudulent based on gesture data (finger location on the screen, acceleration of the phone, finger size, finger pressure, etc.)