Vikram Singh

Vikram Singh

Experience

Experience

Cervest

Software Engineer
Jan 2022 – Present

  • Productionized machine learning models in collaboration with Data Scientists by creating Python libraries and introducing reusable CI/CD workflows using GitHub Actions.
  • Implemented ELT/ETL data pipelines using Argo workflows on Kubernetes (AWS EKS) to accelerate data validation times from 40 hours to 1 hour.
  • Lead the migration of data pipelines for processing multiple TB’s of data to Databricks which uses PySpark’s distributed capabilities.
  • Optimized PySpark Pandas UDF using NumPy to reduce costs by 20 times.
  • Contributed to FastAPI based library for delivering data within a Service Oriented Architecture(SOA) framework.
  • Created Hyperparameter optimization for statistical models, reducing workflow time by 10x.
  • Helped promote best practices by upskilling others using pair programming, proposing improvements in ways of working and collaborating on code guidelines.
  • Contributed to the open source AWS SDK for Pandas.

Max Planck Institute for Meteorology

Scientific Programmer
Jul 2019 – Dec 2021

  • Developed new mathematical model for the in-house climate model that delivered increased precision and accuracy.
  • Built Python based ETL pipelines for visualization of large climate geospatial data output from climate model simulations.
  • Co-ordinated projects with multiple researchers in multiple countries resulting in new publications.

Aeronautical Development Agency

Engineer
Aug 2009 – Oct 2015

  • Developed a Python-based Hyperparameter Tuning framework for aerodynamic design, resulting in new configuration design delivery time reducing from 5 days to 10 hours.
  • Utilized Genetic Algorithms for optimized configuration derivation and achieved significant results that earned me the DRDO Young Scientist Award.

Education

Technion

PhD in Mechanical Engineering
Oct 2015 – Jul 2019

I worked on the Discontinuous Galerkin (DG) method, an arbitrary high-order, unstructured method for the solving the compressible Navier Stokes (CNS) equation. High order methods suffer from aliasing instabilities which leads to robustness issues. I showed that we can solve this issue if we used split forms. In addition, split forms work well with wall modeling to enable high Reynolds’ number flows.

IIT Kanpur

BTech-MTech in Aerospace Engineering
Aug 2004 – May 2009

I worked on airfoil design to optimize for bypass transition. I developed a FORTRAN code for design using B-splines and introduced a new technique to optimize airfoil flow characteristics.