Employer description & source details
As a member of our AI Application team, you'll own getting the latest AI models running fast and accurately on the GPNPU, from CNN perception models to LLMs and VLAs for automotive, robotics and AI-PC. You'll profile real workloads, write and tune kernels, choose quantization strategies, and build the tools that make model deployment repeatable. Analyze and optimize performance on GPNPU Deploy and profile models on GPNPU, from vision/perception to LLM/VLM/VLA Analyze the accuracy and choose the best quantization scheme Write and tune kernels and optimize model compute graphs on GPNPU, balance speed and accuracy Create efficient flows to convert and deploy models on GPNPU Develop efficient tools and flows to speed up the model deployment, including profiling, tracing, visualization tools to find bottlenecks and agentic automation Customize open-source model optimization/quantization tools Integrate and automate model pruning, sparsification, quantization in the workflow Within your first 6-12 months, you’ll: Deploy 5 new models on GPNPU independently, meeting accuracy targets Deliver a 2x speedup on a model through kernel, model compression or dispatch work Ship a deployment or an optimization tool the team uses every day 3+ years experience in AI model deployment or kernel optimization Hands-on performance analysis: profilers, roofline/bandwidth reasoning, finding compute vs memory bottlenecks Hands-on experience with quantization (PTQ/QAT, int8/int4) and debugging accuracy loss Has written performance-critical kernels (CUDA, Triton, OpenCL, SIMD intrinsics or DSP) Fluent with AI coding agents (e.g. Claude Code, Cursor), and interested in building agent-driven workflows for kernel and model bring-up Hands-on experience in pruning and sparsification Solid understanding o
Extracted skills: python, c++