AI Software Engineer - AI Frameworks

Overview

Design and develop optimizations for high-performance AI serving and training frameworks such as SGLang, vLLM, and PyTorch, enabling state-of-the-art solutions for Intel's AI accelerators and next-generation GPUs within the Data Center and AI group (DCAI).

What you'll do

• Design and develop custom GPU kernels for AI framework operations including attention mechanisms, collective operations, and computation primitives to optimize performance on Intel accelerators and GPUs.
• Optimize and enhance framework components including scheduling, memory management, and execution pipeline architecture to improve throughput, latency, and resource utilization.
• Write and integrate high-performance kernels using GPU programming frameworks to maximize throughput and minimize latency.
• Identify and implement optimization opportunities in the framework stack for production AI workloads.
• Collaborate with the open-source community (SGLang, vLLM, PyTorch, and others), participate in upstream development, and integrate new innovations into production software.
• Debug and optimize complex multi-layered systems involving kernel execution, memory hierarchies, and distributed AI scenarios.

What you'll need

• Bachelor's degree (B.Tech) or Master's degree (M.S./M.Tech) in Computer Science, Electronics and Communications Engineering (ECE), or a related field, with 6 to 12 years of professional experience.
• Advanced proficiency in C++ (C++14/17) and intermediate skills in Python and parallel programming.
• Proven, hands-on experience writing and optimizing GPU kernels.
• Demonstrated expertise in kernel development for machine learning operations such as GEMM, attention mechanisms (including Flash Attention), collective operations, and sampling.
• In-depth, hands-on experience with AI frameworks: SGLang, vLLM, PyTorch, or similar production-grade serving and training systems.
• Deep understanding of their internal architecture, scheduling mechanisms, and memory management strategies is essential.
• Strong knowledge of AI inference and training concepts including batching strategies, memory optimization, distributed execution, and performance tuning.
• Practical experience optimizing and deploying AI models at scale in production environments.
• Strong ability to debug complex issues in multi-layered software systems, performance profiling, and bottleneck identification in AI frameworks.
• Solid understanding of computer architecture, GPU execution models, memory hierarchies, and hardware-software co-optimization techniques.
• Experience working on frameworks and platforms that have been deployed to production with high availability and performance requirements.
• Excellent communication skills and experience collaborating with cross-functional, geographically distributed teams.

Nice to have

• Extensive experience developing production-grade kernels with CUDA, CUTLASS, or Triton for various AI operations.
• Published or open-source contributions demonstrating kernel optimization expertise.
• Knowledge of compiler algorithms for heterogeneous systems and kernel fusion optimizations.
• Contributions to or deep familiarity with SGLang, vLLM, PyTorch, or other open-source AI frameworks.
• Experience with distributed AI workloads, multi-GPU scheduling, and optimization across heterogeneous hardware.
• Understanding of quantization techniques, mixed-precision inference and training, and their integration into AI frameworks.

Details

• Location: India, Bangalore.
• Work model: on-site presence required.
• Job type: Experienced Hire.
• Shift: Shift 1 (India).