Lead a team of outstanding researchers and engineers that focuses on research and development on a wide variety of ML domains and HW/SW co-design.

I am the head of the AI team at Cerebras leading the ML Framework and the Applied AI team. Focusing on creating a general and scalable backend for popular deep learning frameworks optimized for Cerebras Wafer-Scale Engine (WSE) and training Large Language & Vision models to SoTA on CS-2.

* Leading a great team of engineers and researchers to drive deep learning framework forward, focusing on PyTorch, ONNX, ONNX Runtime, Horovod and ML.NET.* Participate in all architecture and technical design discussions across the AI framework org.* Influence the organization for our future bet and focus, both in software and hardware. By prototyping and exploring new ideas.* Drive ONNX standard forward inside and outside the company, in addition of leading the operator SIG in ONNX.* Work closely with MSR to evaluate which innovation to incorporate in our pipeline. Collaborate with multiple teams in Microsoft and work closely with our external partners.* Created HyperScale team to focus on high-performance distributed training algorithm and pipeline. We released Adasum distributed algorithm to horovod and became member of Horovod.* Worked with AutoML team for their end-to-end architecture. And delivered the needed components to scale their system.* Mentoring, coaching and helping people to grow both as I.C. and as manager.* Presented our works at multiple conferences.

* Cocreator of ONNX standard which is a common IR for ML and DNN models.* Improving computer vision support in CNTK for both image and video related algorithms and driving the overall architecture forward.* Write and match state of the art computer vision recipes in CNTK.* Design and implemented ONNX exporter/importer for CNTK.* Implemented a deep learning algorithm for emotion recognize from faces, in still images and videos. I led the engineering side and worked on the research side (published 3 papers for this project).* Improved the performance of our runtime.* Led the design and implementation of the emotion recognition APIs.* Set priorities based on our customer and partner team feedbacks, in addition to long-term vision planning of the Emotion tech in MS Cognitive Service and CNTK framework.* Gave numerous presentations, lectures and tutorials on CNTK and emotion recognition, including latest trends and direction in deep learning.

Led a team for a fully articulated real-time hand tracking project on depth sensor and another one on IR sensor, in addition to pose estimation and gesture recognition. Collaborate with Microsoft scientists on the various stages of the hand tracking tech (https://arxiv.org/abs/1604.06195).Owned the depth reconstruction pipeline for Kinect v2 on both Xbox and PC.Collaborate with Microsoft Scientists for some of the techs used in HoloLens.

* One of the architectures of Xbox NUI platform and its interaction with the rest of the OS. I also implemented NUI APIs for depth, IR, color image, segmentation map, fitness, skeleton, and helped other team members to expose their tech through NUI platform.* Technical lead and owner of the 3D Depth reconstruction pipeline in Kinect 2 (LIDAR Sensor), worked on the depth algorithm in collaboration with MSR and MS Hardware team. I designed and implemented the entire pipeline in both SSE and Compute Shader (CS), and met the compute constrain for XBox. Furthermore, I came up with a smoothing filter that preserve edges and reduce noise which outperformed all other smoothing filters that we tried and doesn’t require extra pass.* Green screening: worked on two algorithms for green screening one based on graph cut and the other based on geodesic transform. I came up with a solution that work in real time without any user interaction or initial setup.* Fitness: designed and implemented skeleton based MET estimator for Xbox, which is used by most Xbox Kinect games. The algorithm is designed to handle all activities and not just standard one.* Designed and implemented an Auto Exposure (AE) algorithm optimized for faces and human skin.* Implemented a lean detector algorithm for Xbox One that is in use in some of the launch titles.* Managed and mentored interns, in addition on being a technical lead and mentor for other employees

I was responsible for the AV (Audio\​Video) playback pipeline in Media Center, which includes: 1. Multiple AV graphs to support all our playback scenarios (i.e. Live TV, Recording content, DVD, VCD, and local file playback). 2. Presentation layer that manage rendering the video frame and handle late frames. 3. Managed interop layer for the UI and the capture service.

Worked on the Server side of Microsoft Update.

Designed and implemented a chunk-based algorithm for streaming seamless uninterruptible DVD quality movie over public internet.