Senior ML Inference / GPU Engineer (Full-time, remote from LatAm)
Ruzora places senior LatAm engineers full-time with US startups. We're hiring a Senior ML Inference / GPU Engineer to join a partner company serving models in production. You'll make large models faster and cheaper to run: lower latency, higher throughput, better GPU use.
What you'll do
- Deploy and serve ML and LLM models with vLLM, TensorRT-LLM, Triton or similar
- Profile and optimize inference: batching, KV cache, quantization, speculative decoding
- Write and tune GPU code (CUDA, Triton kernels) where the frameworks fall short
- Run GPU clusters on Kubernetes with autoscaling and cost controls
- Benchmark models and hardware and pick the right setup for each workload
- Work with researchers to take models from notebook to production
What we're looking for
- 5+ years of software engineering, 2+ years on ML infrastructure or model serving
- Strong Python and PyTorch, plus working C++
- Hands-on with at least one inference stack (vLLM, TensorRT-LLM, Triton Inference Server, TGI, ONNX Runtime)
- Understanding of GPU architecture, memory bandwidth and profiling (Nsight, PyTorch profiler)
- Docker, Kubernetes and a major cloud (AWS preferred)
- Excellent written and spoken English (B2 or higher)
Nice to have
- CUDA or Triton kernel development
- Quantization (INT8, FP8, AWQ, GPTQ) and distillation
- Distributed inference (tensor or pipeline parallelism)
- Ray, KServe or similar serving platforms
Stack: Python, PyTorch, CUDA, vLLM, TensorRT-LLM, Kubernetes, AWS
What you get
- Full-time, 100% remote from anywhere in Latin America
- Competitive salary in USD
- Flexible hours with overlap with US time zones
- Professional development budget, health insurance stipend, equipment allowance
Full role details: https://www.ruzora.com/jobs/ml-inference-gpu-engineer
