Senior LLM Inference Engineer / AI Infrastructure Architect - 20159
Veritaz AB · Stockholm · 6 månader – upp till 12 månader · Deadline 2026-11-20
Veritaz AB seeks an experienced **Senior LLM Inference Engineer / AI Infrastructure Architect** to evaluate and optimize LLM-powered applications and agent workflows. The role focuses on identifying performance bottlenecks, resource constraints, and cost drivers while improving token usage, response times, concurrency, and GPU memory utilization. Core responsibilities include implementing techniques such as prompt caching, batching, and quantization, evaluating smaller models for less complex tasks, and conducting load testing and capacity planning. The architect also compares operating models across hosted APIs, cloud, on-premises, and hybrid setups, formulating strategic recommendations for **GPU capacity and scaling** and transferring knowledge to the development team. Applicants must have substantial practical experience with **production deployment and optimization of LLMs**, supported by a strong understanding of how model size, context length, GPU memory, concurrency, latency, and operational costs interact. Hands-on proficiency in Python, Linux, and container-based environments is required, alongside experience using inference frameworks such as vLLM, TensorRT-LLM, or equivalent tools. Experience with AI agents and agentic workflows is advantageous, and additional merit is awarded for expertise in distributed inference, multi-GPU or multi-node scaling, and assessing infrastructure setups from both technical and financial standpoints.
Veritaz is a leading IT staffing solutions provider in Sweden, committed to advancing individual careers and aiding employers in ensuring the perfect talent fit. With a proven track record of successful partnerships with top companies, we have rapidly grown our presence in the USA, Europe, and Sweden as a dependable and trusted resource within the IT industry.
Job Overview
Role Summary
An experienced Senior LLM Inference Engineer / AI Infrastructure Architect
Key Responsibilities
- Analyze existing LLM-powered applications and agent workflows to identify performance bottlenecks, resource constraints, and cost drivers.
- Optimize inference performance with respect to token usage, response times, concurrency, GPU memory utilization, and overall service capacity.
- Evaluate model selection, including the use of smaller models for less complex tasks where appropriate.
- Assess and implement inference optimization techniques such as prompt caching, batching, and quantization.
- Conduct performance measurement, load testing, and capacity planning for AI services.
- Compare operating models including model APIs, self-hosted cloud deployments, on-premises infrastructure, and hybrid alternatives.
- Develop a concrete decision basis for future operating strategy, including recommendations for GPU capacity, scaling approaches, and economically viable usage-volume thresholds.
- Document findings, implemented improvements, and technical recommendations, and transfer knowledge to the development team.
Technical Stack & Requirements
- Production deployment and optimization of Large Language Models (LLMs).
- Strong understanding of the relationship between model size, context length, GPU memory, concurrency, response latency, and operating cost.
- Hands-on experience with Python, Linux, and container-based deployments.
- Experience with LLM inference frameworks such as vLLM, TensorRT-LLM, or equivalent technologies.
- Experience with performance benchmarking, load testing, and capacity planning for AI solutions.
- Practical knowledge of caching, batching, quantization, and GPU-oriented inference optimization.
Qualifications & Experience
The ideal candidate has substantial practical experience deploying and optimizing language models in production environments. They can translate technical analysis and performance results into clear, commercially relevant recommendations and decision material. Experience with AI agents and agent-based workflows is advantageous. Additional merit is given for experience with distributed inference, multi-GPU or multi-node scaling, and evaluating cloud, self-hosted, and hybrid operating models from both technical and financial perspectives.