Julian Weisbord

Hawthorne AI specializes in 3-6 month software projects in Machine Learning, Computer Vision, and IOT. We will scope and design your project, implement custom end-to-end machine learning pipelines on time, and offer post-delivery support and management. Clients Include: Polaris, Intel, Goldstar Events/TTG, AI-Strategy, Macalogic, and Zuya Past Project Examples: Delivered a recommendation engine and pipeline in AWS to recommend live events to millions of users across the U.S based on their preferences and purchase history. Hawthorne AI is building/scaling out 5 ML models for a DoD sponsored, AI enabled decision guidance platform. We have developed a 2 year roadmap and are acting as the Head of Product for their Data Science Team. Developed a model and pipeline to forecast vehicle sales for the next 8 months, taking into consideration supply chain issues. The pipeline also automatically retrains when new data is available and incorporates the challenger model framework   Deployed Fuzzy text matching model using KNN on 5 million+ browser records to identify web users, eliminating need for cookies. The production model identifies users in < 120 milliseconds Created and deployed a Vehicle object detection and license plate OCR system running on Nvidia Jetson Nano’s

Built Automatic ML Pipeline using AWS Sagemaker, consisting of 40+ GB of data cleaning/processing, model training, model optimization, A/B Testing, and deployed event recommendations to 11 million users through our web, mobile, and email services.

• Built a Computer Vision pipeline for training and evaluation that uses YOLOv3, Darknet, Tensorflow, and OpenCV to detect and track objects which enter dangerous areas on industrial job sites. I implemented our core ML services with Tensorflow/Keras and AWS IOT running on Raspberry Pi’s, and streaming incident videos to AWS • Designed and implemented key product features: ► Exclusion Zone Monitoring for People/Vehicles/Equipment ► Model Evaluation ► Personal Protective Gear Tracking ► Edge Device Management

Prototype, productize, and optimize CNN Models & Architectures for 3D human pose & gesture recognition using a laptop webcam to immerse users into the virtual world. Responsibilities: ► Design experiments and tests for various state of the art neural network architectures for classification, object detection/recognition, and 3D human pose estimation tasks using OpenCV + Tensorflow. ► Code algorithms for data pre-processing and post-processing for CNN architectures. ► Setup training pipeline and infrastructure for deep learning on Azure & GCP.

• Data Scientist and Project Manager for a team that applies Deep Learning and DeepMind’s Elastic Weight Consolidation to a Fetch mobile robot for Object Recognition using Robot Operating System (ROS),Tensorflow, Keras, and Python • Designing a Probabilistic Framework for Short-Term Person Recognition using Expectation Maximization, K-means, and Deep Learning with OpenCV and Tensorflow. The robot will recognize people via camera data and provide assistance based on the person’s identity and role. • The goal of these projects is to create an intelligent agent that can recognize objects/people and continuously learn new things about them over time

• Designed and implemented a 12V State Machine Model to make a modular system for cross-team validation of the Model 3 • Model 3 Vehicle Security Controller and NFC Locking Validation • HIL and SIL Software Design Testing • Automated Test Creation in Python and Robot

Projects(Metrology, Computer Science, electrical Engineering): KUKA Robotics, 3D scan automation (Python and JAVA): Using computer vision algorithms and a collaborative robot to mimic how humans scan and generate a CAD file of a part. Python Programming for GOM ATOS ScanBox X-ray /radiography automation(embedded device programming): Deploying a reproducible solution to automate the X-ray process. Adding a microcontroller, programming it in C, and designing an external circuit to parse a barcode that contains part specifications. This solution eliminates most human error. Refactoring Python modules so nontechnical employees can understand them. Technical Writing: Writing automation proposals, timelines, presentations, and managing all aspects of my projects to generate the best ROI.

Roles: Designing electrical systems, 3-D modeling/printing, and making mechanical decisions while maintaining a budget for a network of tablets, which users can access remotely from a Docker container. Users can also access the hardware of tablets which are controlled by Rasberry Pi's. Quality Assurance: running and analyzing different stress tests on Android tablets, writing Android tablet stress tests in Python for an automated robot controlled by an Arduino. Debugging and rewriting python modules for a telemetry system reporting to a CouchDB database.