Dr. Lokesh Ragta

Experience

• 

  Indian Institute of Technology, Delhi

  Jan 2014 - Jul 2017

  ✔ Developed a Cartesian Grid Method (CGM) for UGKS (UGKS-CGM solver) so as to enable UGKS to compute flows past arbitrary shaped geometries.✔ Parallelised the UGKS-CGM solver using MPI by domain division in the physical space.✔ Extended the applications of UGKS to simulate decaying homogeneous isotopic turbulence by using an efficiently parallelised 3D UGKS solver. ✔ Programmed and validated a 3D Unified Gas Kinetic Solver (UGKS) for canonical test cases.✔ Efficiently parallelized the 3D UGKS solver using Message Passing Interface (MPI) and Open Multi-processing (OpenMP).

  • Senior Reseacrh Fellow

    Aug 2016 - Jul 2017

  • Junior Research Fellow

    Jan 2014 - Jul 2016
• 

  Aditya Birla Group

  Aug 2017 - Jun 2018

  Scientist

  1. Worked on the development of a Population Balance Model (PBM) for the full-scale ball mill used in the cement industry. The aim of the project was to determine the Particle Size Distribution (PSD) at the outlet of an industrial scale ball mill. The project was done in the following two stages. a. A PBM model for the lab scale ball mill was developed and validated with the results in the literature. b. The PBM for single stage mill was extended to the full-scale mill by considering the full-scale mill as a series of single-stage mills with the output of one being used as the input of another.2. I was involved in the coupled simulation where we combined Computational Fluid Dynamics (CFD) with the Discrete Element Method (DEM). The aim was to study and predict the accurate flow physics of cyclone. For this purpose, we used Ansys for the CFD part and Rocky/EDEM for the DEM part.3. I was partially involved in the project of optimizing the energy consumption of a ball mill. The aim of this project was to find the optimal ball size that should be used inside the mill so as to minimize the energy without any deteriorating effect on the particle size of the cement. Show less

• 

  STFC

  Sept 2018 - Jul 2020

  CFD and HPC Scientist

  1. Worked on the project titled High-Level Mesoscale Modeling System (HiLeMMS) to develope a domain-specific language for the Lattice Boltzmann Method (LBM) based fluid flow solver. 2. The developed interface allows a user to run his code on multiple computing architectures such as conventional CPUs, multiple CPUs, Graphical Processing Unit (GPUs), Xeon Phis or a combination of these. For this, an underlying library namely Oxford Parallel Library for Structured Mesh Solvers (OPS) is responsible for translating and generating the target-specific source code. This allows a user to focus more on getting the physical insights into a problem rather than learning new programming tools and techniques for each new computing hardware.3. More details can be found at - https://github.com/lkr-ind/MPLB. Show less

• 

  University of Leicester

  Jul 2020 - Feb 2023

  Research Software Engineer Specialist

  1. Accelerated the Classical Gram Schmidt (CGS) orthogonalisation algorithm in the BcDor code by "25X" on Ampere Nvidia GPUs as compared to the LAPACK CPU routines.2. Used Magma library on GPU to accelerate the Eigen value solver in the BcDor by "2X" as compared to CPU.3. Added 3 idealized (Held Suarez, Earth-Like and Tidally Locked Earth) and 2 hot-Jupiter temperature forcing cases (in 3D) to a new Met Office code namely LFRic (a new weather prediction code and a replacement for Unified Model/UM) which is capable of simulating Exoplanets at Exascale.4. Developed post processing codes in Python based on libraries such as Iris, Aeolus (Libraries to analyze and visualize Earth Science Data ) etc. to analyze the results obtained from the Unified Model (UM) and/or LFRic.5. Streamlined and documented the process of installing LFRic and its software stack on DiaL3 using the Intel compilers.6. Used Reframe to create new regression tests for the DiRAC HPC facility and merged them with the ExCALIBUR repository (https://github.com/ukri-excalibur/excalibur-tests).7. Developed the documentation of DIaL3 HPC facility and created a website using MkDocs (https://dial3-docs.dirac.ac.uk).8. Conducted two half day courses to teach the basics of Python and Scientific Python by utilising libraries such as Numpy, Pandas, Matplotlib etc. (https://github.com/UniOfLeicester/PythonCourses).9. Used Arm Forge to profile and benchmark a few astrophysics codes namely: Ramses, SphNG, and Trove on the DiRAC supercomputing facility.10. Prepared and distributed a set of test cases to various vendors namely: HPE, Intel, and AMD for the procurement of DiRAC3 clusters (a part of the work is available at https://github.com/lkr-ind/di-benchmarking-public).11. Developed a Web Scraper based on Selenium (Python) to extract the details of papers that have been published by the DiRAC users in a given period of time and to count the number of year-wise citations of those papers. Show less

• 

  Imperial College London

  Mar 2023 - now

  Senior Research Software Engineer (HPC Consultant)

  1. Accelerated an Image analysis program on Nvidia A100 GPU by "75X" by writing raw kernels in CuPy as compared to the CPU version based on NumPy. 2. Assisted an HPC group in optimising their LAMMPS application by “1.37X” times by using hybrid parallelisation (MPI and OpenMP) and thread pinning.3. Collaborated with Nvidia to organise a workshop for 45 participants on the topic "Fundamentals of Accelerated Computing using Cuda C/C++".4. Developed and conducted a new Unit Testing Course based on the GoogleTest library. More details can be found at https://imperialcollegelondon.github.io/unit_testing_Cpp.5. Conducted multiple workshops for graduate students on several topics such as: Essential Software Engineering for Researchers, Basics of version control using Git and Intermediate Git for effective collaboration.6. Assisted researchers in installing software such as CMAQ, FreeFEM, Nektar++, Hermes-3 etc. in their home directory.7. Supported the HPC users in debugging and optimising their applications and troubleshooting their workflows. Show less