In short
Builds and optimizes inference pipelines for large-scale model serving (LLMs and beyond), working with PyTorch, TensorRT and vLLM, and applying quantization, kernel fusion and hardware-aware optimizations (CUDA/GPU) to tune performance of AI systems.
About the role
- Building and optimizing inference pipelines for large-scale model serving (LLMs and beyond)
- Working with frameworks like PyTorch, TensorRT, and vLLM to deploy models efficiently
- Implementing and optimizing ML models using techniques such as quantization (INT8/FP8), kernel fusion, and efficient batching
- Optimizing and implementing core ML operators (e.g., GEMMs, convolutions, activations, ...)
- Investigating and resolving issues through system-level debugging and performance analysis
- Defining and applying practices for testing, deployment, and scaling AI systems
- BSc/MSc in Computer Science, Engineering, Mathematics, or related discipline
- Strong programming skills in C/C++ or Python in Linux environments using common development tools
- Solid knowledge of computer architecture, system software, data structures
- Hands-on experience implementing algorithms in high-level languages (C/C++/Python)
- Exposure to specialized hardware (GPUs, FPGAs, DSPs, AI accelerators) and frameworks such as OpenCL or CUDA
- Experience designing or working with high-performance software systems
- Solid knowledge of ML fundamentals
- Motivated team player with a strong sense of responsibility
- Experience in at least one of the following areas: Model serving frameworks (e.g., Triton Inference Server, DeepSpeed Inference, vLLM)
- Experience in ML runtimes (e.g., ONNX Runtime, TVM, IREE, XLA)
- Experience deploying ML workloads (LLMs, VLMs, NLP, etc.) across distributed systems
- Experience implementing and optimizing ML operators and kernels with a focus on vectorization and efficient execution (e.g., activation, pooling, quantization)
- Experience in hardware-aware optimizations and performance tuning
- 2+ years of experience developing software targeting AI hardware
- Contribution to open-source projects (e.g., LLVM/MLIR, PyTorch, TensorFlow, ONNX Runtime, xDSL, IREE) is a big plus
Requirements
- BSc/MSc in Computer Science, Engineering, Mathematics, or related discipline
- Strong programming skills in C/C++ or Python in Linux environments