Principal Engineer, Model Performance (AI)

Location: United States; remote

Company

hirify.global builds and operates an AI inference platform and its own distributed infrastructure, including the Sonic Inference Engine and custom inference pods.

What you will do

• Own inference performance across image, video, audio, 3D, and LLM workloads on every GPU generation in the fleet.
• Set the model performance roadmap based on traffic, cost, hardware, and upcoming pod generations.
• Write and tune CUDA kernels, optimize attention and MLP paths, introduce FP8, FP4, and INT4 quantization, and implement multi-GPU parallelism and batching strategies.
• Build the pipeline that takes newly released open models from weights to optimized, production-grade deployments within days.
• Shape the serving stack across TensorRT-LLM, vLLM, SGLang, custom pipelines, and the Sonic Inference Engine; contribute upstream where appropriate.
• Own reproducible, hardware-aware benchmarking and regression detection, partner with infrastructure and hardware teams, and hire, mentor, and lead the Model Performance team.

Requirements

• 10+ years in GPU, HPC, or deep learning systems engineering, including several years optimizing production inference at scale.
• Track record of shipping state-of-the-art models from reference implementations to the Pareto frontier of latency and cost on real hardware.
• Deep hands-on CUDA and GPU architecture expertise, including kernel design, memory hierarchy, occupancy, Tensor Cores, fused kernels, CUDA Graphs, NCCL, Nsight Systems, and Nsight Compute.
• Production experience with a major inference engine such as TensorRT-LLM, vLLM, SGLang, or a similar system, plus continuous batching, paged KV cache, speculative decoding, prefill/decode disaggregation, quantization, and tensor, pipeline, or expert parallelism.
• Fluency in Python and C++; comfort with PyTorch internals, torch.compile, and Triton.
• Experience building or leading engineering teams, hiring, setting technical direction, and remaining hands-on.
• Daily use of AI for kernel search, profiling, benchmarking, documentation, and code review.
• Ability to own latency, cost, and quality outcomes in a fast-moving and ambiguous environment.

Nice to have

• Diffusion and video generation performance experience, including DiT, UNet, long-sequence attention, distillation, caching, LoRA hot-swapping, ComfyUI, Diffusers, or custom engines.
• Contributions to TensorRT-LLM, vLLM, SGLang, FlashAttention, CUTLASS, Triton, Dynamo, or another widely used open-source inference project.
• Experience with Blackwell hardware, NVLink, NVSwitch, InfiniBand, RoCE, and multi-node inference.
• Mixture-of-experts serving, including expert parallelism, placement, load balancing, and memory or communication trade-offs.
• Accuracy evaluation for optimized models, non-NVIDIA accelerators, speech and audio serving, or published inference-performance work.

Culture & Benefits

• Remote-first work with core collaboration hours and flexible scheduling outside them; in-person planning and company retreats take place twice a year.
• Fast release cycles followed by planned downtime.
• Meaningful stock options.
• Generous paid time off, including vacation, sick days, and public holidays.
• Paid maternity, paternity, and caregiver leave.