Position: DevOps Engineer LLM \& GPU Inference Services
Location: Remote
Duration: 1 Years
Skill Rating
Skills Matrix
Skill
Last Used
Experience In Years/month
Rating (10 points)
1 \= newbie 10 \= expert
Hands on Exp.
Yes/No
Cloud environments
Large Language Models (LLMs), particularly hosting them to run inference
Distributed Services Experience
GPU (Dedicated Inference Service)
Job Description
We are looking for devs with general cloud services / distributed services experience, with LLM experience as a secondary skill. GPU experience is now low on the list of preferred skills: Dedicated Inference Service
Required Skills-
Deep experience building services in modern cloud environments on distributed systems (i.e., containerization (Kubernetes, Docker), infrastructure as code, CI/CD pipelines, APIs, authentication and authorization, data storage, deployment, logging, monitoring, alerting, etc.)
Experience working with Large Language Models (LLMs), particularly hosting them to run inference
Strong verbal and written communication skills. Your job will involve communicating with local and remote colleagues about technical subjects and writing detailed documentation.
Experience with building or using benchmarking tools for evaluating LLM inference for various models, engine, and GPU combinations.
Familiarity with various LLM performance metrics such as prefill throughput, decode throughput, TPOT, and TTFT
Experience with one or more inference engines: e.g., vLLM, SGLang, and Modular Max
Familiarity with one or more distributed inference serving frameworks: e.g., llm-d, NVIDIA Dynamo, and Ray Serve etc.
Experience with AMD and NVIDIA GPUs, using software like CUDA, ROCm, AITER, NCCL, RCCL, etc.
Knowledge of distributed inference optimization techniques - tensor/data parallelism, KV cache optimizations, smart routing etc.
What You'll Be Working On-
Develop and maintain an inference platform for serving large language models optimized for the various GPU platforms they will be run on.
Work on complex AI and cloud engineering projects through the entire product development lifecycle (PDLC) - ideation, product definition, experimentation, prototyping, development, testing, release, and operations.
Build tooling and observability to monitor system health, and build auto tuning capabilities.
Build benchmarking frameworks to test model serving performance to guide system and infrastructure tuning efforts.
Build native cross platform inference support across NVIDIA and AMD GPUs for a variety of model architectures.
Contribute to open source inference engines to make them perform better on DigitalOcean cloud.