ML Infrastructure & Acceleration Engineer - LLMs

Who We Are

We are Prosus, a global technology company powering some of the world's leading lifestyle ecommerce businesses. Every day, billions of people across India, Latin America, Europe and beyond use our products and services to buy, sell, pay, connect and discover.

Our AI-driven ecosystem spans 100+ companies across ecommerce, food delivery, fintech and experiences, serving billions of consumers worldwide. AI powers how we operate and innovate, with more than 60,000 active AI agents helping our teams build, scale and move faster. Through our investment team, we also back the founders and technologies shaping the future: From robotics and autonomous systems to synthetic biology and frontier AI.

Guided by our five values Innovation, Results, People, Impact, and Entrepreneurship, culture is our competitive advantage. We thrive with experimentation; we build, we test, we learn, we iterate.

Who we're looking for

We're seeking an experienced ML Infrastructure & Acceleration Engineer - LLMs to own the GPU infrastructure that powers our LLM training and serving, and to make every workload on it faster and cheaper. You'll run our H200 cluster end to end, deploy models to production, and profile and optimize our training and inference pipelines down to the kernel level when it matters. You understand GPUs from the cluster scheduler to the memory hierarchy, you measure before you optimize, and you're motivated by turning raw compute into faster iteration for our ML team. You thrive in environments where infrastructure reliability and efficiency directly shape how quickly research reaches millions of users.

• Amsterdam
• Full time
• AI & ML

What you'll do

GPU Cluster & Infrastructure:

• Own our H200 GPU cluster end to end: Kubernetes and Slurm for training and inference workloads, with multi-tenant scheduling, quotas, and resource isolation
• Keep large distributed training runs healthy: node health checks, NCCL and interconnect debugging, automated job recovery, and checkpoint management
• Track and optimize GPU utilization and cost across training and inference, with dashboards and budget alerting
• Manage infrastructure-as-code and provisioning for GPU environments
• Define and monitor SLIs/SLOs for cluster availability and production model serving

Deployment & Serving:

• Deploy LLMs to production using vLLM or SGLang, with CI/CD, model versioning, and safe rollouts and rollbacks
• Monitor serving latency, throughput, errors, and cost, and act on regressions

ML Acceleration:

• Profile training and inference end to end (PyTorch Profiler, Nsight Systems, Nsight Compute) to find and remove bottlenecks
• Improve training throughput and MFU: parallelism strategies (FSDP, tensor and pipeline parallelism), communication/compute overlap, mixed precision and FP8, activation checkpointing, torch.compile
• Optimize inference: quantization (FP8, INT8, INT4), speculative decoding, prefix caching, continuous batching, KV cache tuning, and parallelism layouts
• Write or integrate fused and custom kernels (Triton, CUDA) where off-the-shelf implementations fall short
• Build benchmarking tooling so every optimization ships with measured gains

Enablement & Best Practices:

• Partner with ML engineers and researchers to make training and serving efficient by default
• Establish best practices for using the cluster and guide teams in their adoption

Minimum qualifications

• 5+ years in ML infrastructure, MLOps, HPC, or platform engineering with focus on GPU workloads
• Hands-on experience operating multi-node GPU clusters with Kubernetes and/or Slurm, including debugging hardware, networking (NCCL, InfiniBand/RoCE), and job failures
• Production experience deploying and optimizing LLM inference with vLLM, SGLang, or TensorRT-LLM
• Proven track record of profiling and accelerating GPU workloads, with measured results in throughput, latency, MFU, or cost
• Solid understanding of monitoring and observability for GPU systems (Prometheus, Grafana, DCGM, or similar)
• Naturally curious with a track record of proactively identifying and implementing improvements

Preferred qualifications

• Experience writing custom kernels in Triton or CUDA, and familiarity with fused kernels such as FlashAttention
• Performance tuning of distributed training frameworks: FSDP, Megatron-LM, or DeepSpeed
• Experience with low-precision training and inference (FP8, INT4) and model compression for production deployment
• Deep knowledge of GPU architectures: memory hierarchy, tensor cores, roofline analysis
• Experience with multi-tenant GPU scheduling and fair-share resource allocation
• Contributions to open-source projects such as vLLM, SGLang, PyTorch, or Triton

What we offer

• Critical infrastructure ownership for high-impact AI projects that are strategically vital to the company, with direct visibility to senior leadership including the CEO
• State-of-the-art GPU infrastructure: H200 fleet, vLLM serving stack, and the freedom to optimize it end to end
• Expert ML team who have released top Hugging Face models, published at NeurIPS, and built production systems that will run on your infrastructure
• Significant autonomy in choosing tools and shaping our infrastructure and performance strategy
• Modern tooling: Latest ML frameworks, coding assistants, best-in-class development environment
• Hybrid work model with our Amsterdam office - home to the AI House, bringing together 200+ AI professionals through events and collaborations
• Competitive compensation, top-spec MacBook Pro, and an environment genuinely built for professional growth and learning

If you're excited to own the GPU infrastructure behind cutting-edge LLMs and make every training run and every request faster, let's talk.