Mohamed Ezzeldin, Machine Learning Developer in Cambridge, MA, United States
Mohamed Ezzeldin

Machine Learning Developer in Cambridge, MA, United States

Member since December 28, 2018
Mohamed is a machine learning tech lead at Affectiva. He has worked on face detection, facial landmark tracking, head pose estimation, and object detection. Previously, he worked as an embedded and back-end software engineer and a machine learning researcher (Computer Vision). Mohamed skillfully combines technology to set up ML cloud pipelines and optimize ML models to run on mobile and embedded devices.
Mohamed is now available for hire

Portfolio

  • Affectiva
    Python, TensorFlow, TensorFlow Lite, Keras, Docker, AWS, C++, Qualcomm SNPE...
  • Cliqz
    Python, AWS, Elasticsearch, Redis, Flask
  • Microsoft
    C++, Visual Studio

Experience

Location

Cambridge, MA, United States

Availability

Part-time

Preferred Environment

Python, TensorFlow, Keras, Pandas, Scikit-learn, TensorFlow Lite, NumPy

The most amazing...

...migration I've run was Affectiva's entire research department from classical ML to Deep NNs, including building data, training pipelines, and hiring the team.

Employment

  • Machine Learning Tech Lead

    2016 - 2020
    Affectiva
    • Migrated Affectiva's research from classical ML approaches to Deep NNs by building a concurrent data preparation and model training framework and setting up on-premise and cloud infrastructures to help researchers spend more time on research.
    • Trained and productized substantially improved models for face detection, facial landmark tracking, head pose estimation, and object detection.
    • Led the research and engineering efforts to port the entire face analysis pipeline to TFLite and Qualcomm SNPE run times.
    • Led PoC for multiple features such as in-car body pose estimation, infant-seat, and mobile phone detection.
    • Helped hire a number of Computer Vision researchers with emphasis on Deep Learning expertise.
    • Translated customer requirements into data and research needs, including specifying the amount and specs of data, researcher time, and assessing overall risk and achievable KPIs.
    • Organized a leading meetup in Cairo about Deep learning and Computer Vision in Automotive (Cairo AI Meetup). Presented Affectiva in multiple technical events and summer schools in Egypt and Germany.
    Technologies: Python, TensorFlow, TensorFlow Lite, Keras, Docker, AWS, C++, Qualcomm SNPE, Jenkins
  • Back-end Software Engineer

    2014 - 2016
    Cliqz
    • Built a news articles indexing and keyword search pipeline.
    • Developed an article search ranking system based on social media signals such as reactions, retweets, and shares.
    • Optimized search response time in millions of news articles to below 10 milliseconds.
    Technologies: Python, AWS, Elasticsearch, Redis, Flask
  • Software Engineer Intern

    2015 - 2015
    Microsoft
    • Added Microsoft Office support for the new virtual desktop feature of Windows 10. Word was the first MS Office app to support this feature.
    • Initiated discussions with engineering managers of PowerPoint and Excel to roll out virtual desktop support across MS Office apps.
    • Developed a set of reusable functionalities in C++ MS Office code base to provide Win10's virtual desktop capabilities to the rest of the Office product suite.
    Technologies: C++, Visual Studio
  • Back-end Software Engineer

    2013 - 2014
    Affectiva
    • Developed substantial features in a video labeling tool (web) and made optimizations to allow increasing the number of concurrent labelers/users to scale up labeling operations.
    • Reduced manual data labeling cost by 2-3x by implementing an ActiveLearning pipeline and using ML classifiers to select promising videos to label.
    • Built an ETL pipeline to feed the largest repository of human facial expressions and improved database schema of the database for real-time data mining using a distributed columnar database (Vertica).
    • Improved frequently-used query times by more than eight times by finding and removing long standing bottlenecks in the distributed database cluster.
    Technologies: Python, Django, MySQL, JavaScript, Vertica, AWS
  • Embedded Software Engineer

    2011 - 2012
    Valeo
    • Maintained a Hardware Abstraction Layer (drivers) for Park4U (auto-parking) on a multi-core PowerPC micro-controller including a driver for CAN Bus.
    • Created GUI-configurable automated tests for several C modules in Park4U.
    • Developed a static analysis tool to test the requirement compliance of executables based on compiler (windriver) output reports.
    Technologies: C, C++, Qt, DOORS, PowerPC, WindRiver

Experience

  • Saratan: Automatic Liver and Lesion Segmentation in 3D Medical CT Images (Development)
    https://github.com/mohamed-ezz/saratan

    This project is the official implementation for our paper "Automatic Liver and Lesion Segmentation in CT Using Cascaded Fully Convolutional Neural Networks and 3D Conditional Random Fields."
    It implements training, inference, and evaluation pipelines for our algorithm to segment lesions (cancer tissue) in the liver, using convolutional networks and conditional random fields.

Skills

  • Languages

    Python, SQL, C++
  • Libraries/APIs

    TensorFlow, Keras, Pandas, OpenCV, NumPy
  • Platforms

    Amazon Web Services (AWS), Docker, Linux, AWS EC2
  • Other

    Tensorflow Lite, Computer Vision, Artificial Intelligence (AI), Machine Learning, Deep Learning
  • Tools

    Jira, AWS EBS, AWS Batch
  • Paradigms

    Agile, Scrum
  • Storage

    MySQL, AWS S3, AWS RDS

Education

  • Master of Science degree in Machine Learning
    2014 - 2016
    Technische Universit√§t M√ľnchen - Munich, Germany
  • Bachelor of Science degree in Computer Science
    2006 - 2011
    German University in Cairo - Cairo, Egypt

To view more profiles

Join Toptal
Share it with others