Job description for Staff ML Inference Engineer at Cloudera
At Cloudera, we empower people to transform complex data into clear and actionable insights. With as much data under management as the hyperscalers, we're the preferred data partner for the top companies in almost every industry. Powered by the relentless innovation of the open source community, Cloudera advances digital transformation for the world’s largest enterprises.
As adoption of AI grows across our platform, customers are deploying models for increasingly diverse business workloads across on-premises, private, sovereign, and public-cloud environments. We are seeking a Staff ML Inference Engineer to develop and optimize production model-serving solutions through Cloudera AI Inference.
This is a hands-on ML engineering role focused on inference and performance. You will investigate workload requirements, benchmark model-serving configurations, identify bottlenecks, and develop validated solutions across different GPU environments. Working alongside platform and applied AI engineers, you will translate customer requirements into reliable, efficient model endpoints and turn technical findings into reusable engineering capabilities and product improvements.
AI Inference & Model Serving: Deploy and optimize production model-serving solutions through Cloudera AI Inference across on-premises, private-cloud, and public-cloud environments. Validate model, runtime, and GPU compatibility, alongside API behavior, concurrency, and reliability.
Inference Benchmarking & Capacity Planning: Develop repeatable benchmarks using representative input lengths, output lengths, request rates, and concurrency. Recommend GPU types, counts, and serving configurations based on model characteristics, memory requirements, latency targets, throughput, and availability needs.
Inference & Performance Engineering: Diagnose bottlenecks across serving runtimes, GPU compute and memory, CPU processing, networking, and I/O. Evaluate supported optimization techniques such as batching, KV-cache management, parallelism, quantization, and replica scaling, validating their impact on performance and model quality.
GPU Resource Efficiency: Evaluate inference workloads across different GPU generations and configurations. Improve resource utilization while balancing latency, throughput, cost, and operational constraints, including GPU sharing or partitioning where appropriate.
Inference Tooling & Validation: Build benchmarking, profiling, diagnostic, and validation tooling to reproduce customer scenarios and measure time to first token, inter-token latency, end-to-end latency, throughput, and GPU memory consumption.
Data lowongan bersumber dari glints. Tombol “Lamar” mengarahkan Anda ke halaman aslinya.
Product & Engineering Collaboration: Work with customer teams, platform engineers, applied AI engineers, and core engineering to resolve serving challenges. Translate findings into product requirements, performance improvements, tested configurations, and reusable deployment guidance.
Eight or more years of engineering experience spanning ML systems, software development, distributed systems, or performance engineering, with evidence of technical leadership and independent ownership.
Practical experience operating GPU-based model-serving workloads in production, using at least one inference stack such as NVIDIA NIM, vLLM, SGLang, TensorRT-LLM, Triton, or equivalent.
The ability to turn ambiguous requirements into testable hypotheses, investigate root causes, and validate solutions through experiments, profiling, and measurement.
Strong programming skills in Python and a solid foundation in Linux, containers, distributed services, and Kubernetes, enabling you to investigate issues across the serving stack.
Demonstrated experience benchmarking and improving inference performance, with an understanding of model weights, runtime overhead, KV-cache requirements, and latency-throughput trade-offs.
Experience sizing GPU resources and serving configurations against workload demand, including context lengths, output lengths, concurrency, traffic bursts, and reliability requirements.
Experience building reusable inference benchmarks, deployment automation, or diagnostic tools for enterprise environments, including restricted or air-gapped deployments.