Lead Kernel Engineer/Architect (m/f/d)

EPAM Systems; Berlin; Hybrid; EUR 140.000 - 190.000; Vollzeit

Benefits dieser Stelle

30 days holiday per annum

Company Pension Scheme

Regular performance assessments

Discount on Fitness-First Black

ESPP

Learning opportunities and courses

Friendly team and social events

Zusammenfassung

EPAM Systems in Berlin is seeking a Lead Kernel Engineer/Architect to shape AI performance for large-scale training and inference on TPUs and GPUs. You will drive kernel-level optimization and build robust performance tooling in a hybrid work model.

You’ll collaborate with researchers, compiler teams and ML framework developers to deliver high-performance, scalable solutions across cloud platforms, OSS, and production environments.

Qualifikationen

• Bachelor’s degree or equivalent practical experience.
• 12+ years of industry experience in software engineering or systems programming.
• 5+ years of experience in software development using C++ or Python.
• 3+ years of experience in testing, maintaining or launching software products and at least 1 year in software design or architecture.
• Hands-on experience in performance optimization at the kernel level for accelerators or high-performance systems.

Aufgaben

• Design and optimize high-performance kernels for TPU and GPU architectures using low-level programming frameworks such as Pallas, Triton or Mosaic.
• Build and maintain performance infrastructure, including benchmarking suites, autotuning systems, regression testing frameworks and tooling.
• Collaborate with ML framework developers (e.g., JAX, PyTorch) and compiler teams (XLA/MLIR) to integrate custom kernels and reduce performance bottlenecks.
• Track advancements in accelerator hardware, compiler technology and AI model design to identify opportunities for kernel-level optimization.
• Develop clear documentation, APIs and supporting OSS components that improve developer usability and adoption.
• Analyze and resolve complex performance issues impacting large-scale distributed training and inference systems.

Kenntnisse

C++

Python

Kernel optimization

Performance optimization

Ausbildung

Bachelor's degree or equivalent practical experience

Jobbeschreibung

We're looking for a Lead Kernel Engineer/Architect to join our team in Germany in a hybrid working mode.

Are you passionate about pushing advanced hardware accelerators to their limits? Join us in shaping the future of AI performance and scalability.

As a Lead Kernel Engineer/Architect, you will drive the optimization of critical machine learning operations for large-scale training and inference, working with cutting-edge hardware like TPUs and GPUs, advanced ML models and performance toolchains. Your work will enable faster AI research and production deployments on cloud platforms and within open-source ecosystems.

In this role, you will collaborate with researchers, compiler engineers and framework developers to deliver optimized, high-performance solutions that set the standard for modern AI computation.

Responsibilities

• Design and optimize high-performance kernels for TPU and GPU architectures using low-level programming frameworks such as Pallas, Triton or Mosaic
• Build and maintain performance infrastructure, including benchmarking suites, autotuning systems, regression testing frameworks and tooling
• Collaborate with ML framework developers (e.g., JAX, PyTorch) and compiler teams (XLA/MLIR) to integrate custom kernels and reduce performance bottlenecks
• Track advancements in accelerator hardware, compiler technology and AI model design to identify opportunities for kernel-level optimization
• Develop clear documentation, APIs and supporting OSS components that improve developer usability and adoption
• Analyze and resolve complex performance issues impacting large-scale distributed training and inference systems

Requirements

• Bachelor’s degree or equivalent practical experience
• 12+ years of industry experience in software engineering or systems programming
• 5+ years of experience in software development using C++ or Python
• 3+ years of experience in testing, maintaining or launching software products and at least 1 year in software design or architecture
• Hands-on experience in performance optimization at the kernel level for accelerators or high-performance systems

Nice to have

• Proficiency in low-level accelerator programming (CUDA, Triton, Pallas)
• Familiarity with ML frameworks such as JAX or PyTorch and optimization techniques for attention layers, Mixture of Experts (MoE) and precision tuning
• Strong understanding of modern hardware accelerators, including pipelining, data movement and heterogeneous compute
• Knowledge of compiler principles and intermediate representations (e.g., MLIR, OpenXLA)
• Experience building OSS developer infrastructure, APIs and performance-critical libraries
• Excellent problem-solving skills and ability to collaborate in cross-functional engineering environments

We offer

• 30 days holiday per annum
• Company Pension Scheme
• Regular performance assessments
• Discount on Fitness-First Black Membership
• bitkom - Corporate Benefits
• Employee Stock Purchase Plan (ESPP) (subject to certain eligibility requirements)
• Learning and development opportunities, including in-house training and coaching, professional certifications, and courses
• Friendly and enjoyable working team
• Regular corporate and social events
• Flexible and remote working opportunities
• Award-winning workplace: Great Place To Work certified in 2026, Kununu (Top Company 2022-2026), NewWork Business Award 2025 for outstanding culture, innovation and employee satisfaction.