Sergei Bastrakov

Tech Lead AI based in Germany

Email / GitHub / Google Scholar / LinkedIn

Tech lead AI working on a GPU cloud. Former computational scientist for 10+ years, have been core developer and maintainer of particle-in-cell code PIConGPU preparing it to run on world’s first exascale supercomputer. Very experienced in AI, HPC, GPU computing, modern C++. PhD in computer science and a solid background in math.

👨‍💻 Professional Experience

• Technical lead AI @ Northern Data / Taiga Cloud (Feb 2024 - currently)
  • Leading AI team
  • Building AI services for the Cloud
• HPC systems engineer -> Principal HPC engineer @ Northern Data / Taiga Cloud (Dec 2022 - Jan 2024)
  • Setting up HPC and AI workflows as internal customer of the Cloud
  • Developing, running and analyzing performance benchmarks
  • Designing performance metrics and monitoring
  • Investigating and fixing performance-related issues
• Postdoctoral researcher @ Helmholtz-Zentrum Dresden-Rossendorf (Jan 2018 - Sep 2022)
  • A core developer and maintainer of C++ exascale particle-in-cell code PIConGPU
  • Implemented and extended core particle-in-cell routines, including state-of-the-art numerics for incident lasers and field absorbers
  • Prepared and tuned PIConGPU to run on exascale supercomputers
  • Largely contributed to refactoring and modernization of code base and documentation
  • Maintained and contributed to underlying parallel programming libraries alpaka and cupla
• Software developer @ University of Nizhni Novgorod (March 2016 - Dec 2017)
  • Further led development of C++ particle-in-cell code PICADOR
  • Oversaw code architecture, assisted other developers, co-supervised students
  • Designed performant parallel primitives for expression of core particle-in-cell operations and extensions
  • Enhanced concepts and implementation of PIC-MDK interface for portable particle-in-cell extensions, applied it for advanced QED extensions
  • Led development of a new kernel library for particle-in-cell simulations based on experience gained in PICADOR
• Junior researcher @ University of Nizhni Novgorod (Jul 2012 - Dec 2015)
  • Led development of particle-in-cell code PICADOR
  • Implemented advanced particle-in-cell numerics, including charge-conserving current deposition, incident lasers, absorbing boundaries
  • Developed a suite of integration and system tests, set up nightly builds
  • Contributed to initial design of PIC-MDK interface for portable particle-in-cell extensions, implemented it for PICADOR
  • Co-developed a novel particle-in-cell implementation for Intel Xeon Phi
• Part-time Assistant @ University of Nizhni Novgorod (Nov 2010 - Jun 2012)
  • Participated in initial development of particle-in-cell code PICADOR
  • Largely contributed to initial design, implemented most core particle-in-cell numerics
  • Led heterogeneous implementation with OpenCL, that was among the first published fully GPU-capable particle-in-cell codes
• Summer school intern @ Intel (Jul 2008 - Aug 2008; Jul 2009 - Aug 2009)
  • Implemented and optimized basic financial math primitives for option pricing (Jul 2008 - Aug 2008)
  • Implemented a broad phase collision detection algorithm based on octrees (Jul 2009 - Aug 2009)

🏆 Accomplishments

• Scholarship for outstanding young scientists (2016 - 2017)
• Best paper award at Russian Supercomputing Days 2015
  • For a performance study (in Russian) of realistic particle-in-cell applications on Xeon Phi
• Razuvaev scholarship for PhD students (2013 - 2015)
• Contributions featured in press:
  • AMD on PIConGPU simulations on Frontier supercomputer - I have been a core developer of PIConGPU and underlying portability library alpaka
  • AMD on preparing PIConGPU for Frontier supercomputer - I have been a core developer of PIConGPU and worked on features needed for large-scale simulations on Frontier
  • The register on 100 GLOPS/s performance of PICADOR on Xeon Phi - I have been lead developer of PICADOR and co-developed implementation for Xeon Phi
  • R&D world on large-scale PICADOR simulations on Xeon Phi enabling studies of nonlinear vacuum - I have been lead developer of PICADOR and co-developed implementation for Xeon Phi, also co-developed QED-PIC extensions featured

💬 Languages

• English: Advanced, ≈ C1
• Russian: Native
• German: Beginner, ≈ A1 - A2

👨‍🏫 Teaching Experience

Part-time in parallel to other job commitments, all @ University of Nizhni Novgorod:

• Assisted on linear programming class as part of PhD studies (Jan 2014 - May 2014)
• Co-developed a suite of test problems and automatic verification system for parallel numerical methods class (Jan 2013 - May 2014)
• Taught algorithms and data structures for international students in English (Sep 2013 - Dec 2017)
• Assisted on introduction to C++ programming class for international students in English (Sep 2012 - May 2013)
• Textbook contributions (all in Russian):
  • Authored computational geometry chapter of a 2018 textbook on parallel programming problems
  • Authored computer arithmetic chapter of a 2013 textbook on parallel numerical methods
  • Co-authored CUDA chapter of a 2010 textbook on programming for modern architectures

👨‍🎓 Education

• PhD in Theoretical Computer Science (Russian candidate of science)
  @ University of Nizhni Novgorod - Nizhni Novgorod, Russia (Sep 2012 - Oct 2016)
  • My PhD studies dealt with polyhedral computation in general dimension case, namely with incremental algorithms for vertex and facet enumeration to compute dual description of convex polyhedra.
  • Developed a new modification of the double description method combining it with some features of the Quickhull algorithm and obtained complexity bounds for the new algorithm.
  • Proposed an algorithmical improvement to Fourier-Motzkin elimination - a faster way of checking conditions for Chernikov rules / Imbert’s acceleration theorems.
  • For a related problem of incrementally removing vertices/constraints considered a special case and proposed a new algorithm based on reverting iterations of the double description method and proved its polynomiality for this case.
  • Thesis summary (in Russian).
  • Was awarded several scholarships for academic excellence
• MSc in Applied Mathematics and Computer Science with excellence
  @ University of Nizhni Novgorod - Nizhni Novgorod, Russia (Sep 2010 - Jun 2012)
  
  • Diploma work on implementation and properties of general dimension Quickhull algorithm
  • Was awarded several scholarships for academic excellence
• BSc in Applied Mathematics and Computer Science with excellence
  @ University of Nizhni Novgorod - Nizhni Novgorod, Russia (Sep 2006 - Jun 2010)
  
  • In 2007 - 2010 also did extracurricular studies and student projects in parallel programming @ ITLab

♟️ Hobbies

• Table tennis - recreational but passionate club player
• Competitive card game player in Flesh and Blood, previously in Legends of Runeterra, Mythgard and Artifact
• Online chess