Ankur Sakhuja

Software Engineer

Greater Delhi, Delhi, India13 yrs experience
Highly Stable

Key Highlights

  • Expert in deep learning and GPU programming.
  • Proven track record in 3D graphics and visualization.
  • Strong background in Android OS development.
Stackforce AI infers this person is a highly skilled software engineer specializing in AI/ML and graphics software development.

Contact

Skills

Core Skills

Deep LearningGpu ProgrammingComputer Vision3d GraphicsSoftware DevelopmentAndroid DevelopmentEducationDriver DevelopmentValidation EngineeringWeb Development

Other Skills

Performance OptimizationConvolutional Neural NetworksRecurrent Neural NetworksMATLABPythonData AnalysisOpenGLC++QtVisualization ToolkitDebuggingSystem SoftwareTeachingCurriculum DevelopmentProgramming

Experience

13 yrs
Total Experience
1 yr 5 mos
Average Tenure
--
Current Experience

Qualcomm

Staff Engineer

Jul 2017Jun 2021 · 3 yrs 11 mos · Bengaluru Area, India

  • Development and performance tuning of a stand-alone, low-level and multi-platform deep neural network (DNN) inference engine for GPU-based acceleration of deep learning algorithms leveraging convolutional (CNN) and recurrent (RNN) neural networks for on-device machine learning.
  • Protoyping and benchmarking cutting-edge DNNs from recent deep learning research literature (image classification, object detection, semantic segmentation) in order to establish competitive performance of our core library.
  • Performance optimization by exploiting Qualcomm Adreno GPU's special (proprietary) hardware architecture and instructions for graphics and compute API acceleration.
Deep LearningGPU ProgrammingPerformance OptimizationConvolutional Neural NetworksRecurrent Neural Networks

Indian institute of technology, delhi

2 roles

Senior Research Fellow

May 2016Feb 2017 · 9 mos · New Delhi Area, India

  • Appointment in department of Electrical Engineering (through FITT, IIT Delhi) in research project funded by Cadence Design Systems. Worked under CTO office in the IP group (based primarily in San Jose, CA) that pursues the objective of driving technical innovation in deep learning for Tensilica Vision DSPs for imaging and computer vision. Main contributions included (but not limited to) code enhancements, prototyping of ideas from recent deep learning research literature and providing support functions for following:-
  • > Performed training, validation and testing of convolutional neural networks for (a) object detection in images and (b) semantic segmentation of road scenes (an essential component of advanced driver assist systems).
  • > Explored novel approaches for improving correct detection rate of CNN through training on ground truth images in non-spatial domains. Performed analysis of these approaches by computing statistical parameters including image energy distribution, image reconstruction accuracies and run-time computational complexity over standard road image data sets including KITTI and Cityscapes.
  • > Generation of ground truth image data for CNN training and validation through a combination of scripting, collection from industry standard data sets including ImageNet, Microsoft Coco, KITTI, Cityscapes and through manual annotation.
  • Development environment: MATLAB, MatConvNet, Cadence's proprietary deep learning framework, Python.
Deep LearningConvolutional Neural NetworksMATLABPythonData AnalysisComputer Vision

Senior Project Fellow

Aug 2014Aug 2015 · 1 yr · New Delhi Area, India

  • Main contributions and responsibilities include following:-
  • Developed a 3D graphics visualization engine enabling mouse-controlled viewing of Finite Element Model (FEM) of human body. Key features implemented: 3D rotate, zoom and pan of model mesh with a coloring scheme based on anatomical parts and a wireframe overlay. Mouse-picking of mesh nodes based on ray-casting method.
  • Optimized feature performance by enhancing OpenGL command sequence for better ordering of model data in vertex buffers and minimizing computationally expensive OpenGL function calls for rendering.
  • Implemented repositioning module for performing mesh deformation for standard movements of anatomical parts in vehicular driving scenarios. Algorithm used was based on Delaunay tetrahedralization of 3D geometry enclosing the mesh.
  • Optimized repositioning algorithm by implementing strategy for minimizing search space of tetrahedrons for mapping mesh nodes. Efficiency was improved significantly and run-time reduced from several hours to seconds for entire mesh containing millions of nodes and elements.
  • Developed a graphical toolbox module for GUI-based management of geometrical primitives including 3D points, point sets, axes, coordinate reference frames and 3D planes.
  • Implemented parser for reading/writing model data from/to Visualization Toolkit's (VTK) XML-based file format using the open-source TinyXML-2 library.
  • Ported part of code base from legacy implementation (MS Visual C++ with MFC) to a next-generation version using Qt framework for enabling platform-independence in the application.
  • Provided software engineering knowledge transfer to team members from Mechanical Engineering domain, and technical mentorship to junior project members and interns.
  • Development platforms / technologies used: MS Visual C++ with MFC (under MS Visual Studio 2012), C++ Boost libraries, OpenGL, Qt, XML, CMake, Subversion, Git.
3D GraphicsOpenGLC++QtVisualization ToolkitSoftware Development

Samsung electronics

Lead Engineer

Nov 2012Aug 2014 · 1 yr 9 mos · Noida Area, India

  • Software development and debug of system-related issues in Android OS upgrades on Samsung mobile devices.
Android DevelopmentDebuggingSystem Software

Cluster innovation centre, university of delhi

Assistant Professor of Computer Science

Jun 2012Oct 2012 · 4 mos · New Delhi Area, India

  • Ad-hoc appointment as member of faculty in four year undergraduate program "B.S. Innovation with Mathematics & IT" (URL: http://cic.du.ac.in). Courses taught include: data structures, algorithms, programming in C/C++, Prolog.
TeachingCurriculum DevelopmentProgrammingEducation

Advanced micro devices

Software Development Engineer II

Jan 2010Dec 2011 · 1 yr 11 mos · Boxborough, MA

  • DirectX device driver development for ATI graphics processing units.
Driver DevelopmentDirectXSoftware Engineering

Lime brokerage

Real-time Software Engineer

Nov 2008Jan 2010 · 1 yr 2 mos · Waltham, MA

  • Development of software infrastructure for test automation of Lime Brokerage Trading Server - a high-throughput, low-latency electronic trading platform.
Test AutomationSoftware Development

Nvidia

System Software Intern

May 2008Oct 2008 · 5 mos · Santa Clara, CA

  • Software development, debug and testing of NVIDIA GPU graphics drivers on Apple Mac platforms.
Software DevelopmentDebugging

Intel

Sofware Engineer

Aug 2005Dec 2006 · 1 yr 4 mos · Bangalore, India

  • Validation engineering and test automation for Intel graphics device drivers.
Validation EngineeringTest Automation

Ge medical systems

Design Engineer Intern

Jan 2005Jun 2005 · 5 mos · Bangalore, India

  • Design, development and testing of Magnetic Resonance Imaging (MRI) software.
Software DevelopmentTesting

Tata consultancy services

Intern

Jun 2002Jul 2002 · 1 mo · Gurgaon, India

  • Design and implementation of website-search and hit-counter functionality on corporate intranet website of business unit.
Web Development

Education

Boston University

Master of Science - MS — Electrical Engineering

Jan 2007Jan 2008

International Institute of Information Technology Bangalore

Master of Technology - MTech — Information Technology

Jan 2003Jan 2005

Guru Gobind Singh Indraprastha University

Bachelor of Technology - BTech — Information Technology

Jan 1999Jan 2003

Delhi University

Bachelor of Computer Applications

Jan 1999Jan 1999

Springdales school

AISSCE — Science

Jan 1998Present

Stackforce found 100+ more professionals with Deep Learning & Gpu Programming

Explore similar profiles based on matching skills and experience