Вакансия из Hirify Global, списка международных tech-компаний Для мэтча и отклика нужен Plus
Пожаловаться
88
Крутая вакансия
развернуть
В Company hidden предлагают стажировку с понятными задачами в быстро развивающейся области AI-инфраструктуры: работа с реальным оборудованием, бенчмарками и системами, результаты которой могут быть опубликованы.
Современный стек
Перспективное направление
Опыт исследований
?Только для студентов
Мэтч & Сопровод
Для мэтча с этой вакансией нужен Plus
Подключить Plus
Описание вакансии
Текст:
Hirify AI/Оригинал
TL;DR
AI Software Development Intern (Agentic AI): Building and optimizing systems, runtimes, benchmarks, and distributed pipelines for agentic AI on heterogeneous CPU-GPU platforms with an accent on LLM inference, tool execution, sandboxing, and hardware-aware performance. Focus on designing adaptive routers, scaling sandboxed rollouts, measuring CPU-GPU behavior, and evaluating systems on real hardware and clusters.
Location: Hybrid role based in Gdansk, Poland, with time split between the assigned hirify.global site and off-site work.
Minimum annual salary: 68,880 PLN. Final compensation may include additional pay components and benefits.
Company
hirify.global develops processors, accelerators, platforms, and software for data centers, cloud infrastructure, and AI systems.
What you will do
Accelerate CPU-side agent workloads, including tool execution, retrieval, embedding, reranking, and small-model inference, using hirify.global AMX and other accelerators.
Design adaptive routers for model and tool calls based on load, deadlines, cost, and quality.
Build scalable orchestration and execution pipelines for agentic inference and reinforcement-learning-based agent training with concurrent sandboxed rollouts.
Improve sandbox creation and execution for code-executing agents.
Develop CPU-GPU benchmarks and characterization methods covering utilization, contention, queueing, and agent workflow performance.
Build and evaluate proofs of concept on real hardware and clusters using reproducible and statistically sound measurement.
Requirements
Currently enrolled in a Bachelor's, Master's, or PhD program in Computer Science, Computer Engineering, Electrical Engineering, or a related field.
Project, research, or coursework experience in AI/ML systems, LLM inference and serving, agentic AI systems, distributed systems, or computer architecture and performance engineering.
Understanding of LLM serving and inference optimization, including batching, KV caching, or speculative decoding.
Experience evaluating LLM and agent performance, including tool use and code execution, with benchmarks such as llmbench, Terminal-Bench, or SWE-bench.
Working knowledge of AI frameworks and serving stacks such as PyTorch, vLLM, SGLang, LangChain, or LangGraph.
Strong programming skills in Python and/or C/C++, plus proficiency with Linux.
Excellent verbal and written communication, strong ownership, and the ability to take open-ended research problems from idea to working system.
Rigor and honesty in measurement, including willingness to report unexpected results.
Nice to have
Open-source contributions to AI projects such as vLLM, SGLang, Ray, or PyTorch.
Interest or experience in CPU performance, hardware-aware optimization, profiling tools such as perf, or vector instructions such as AVX-512 and AMX.
Working knowledge of observability tools such as Langfuse and Grafana.
Familiarity with containers, virtual environments, and virtualization technologies such as Docker, Python venv/conda, KVM, or VMware.
Familiarity with AWS, GCP, or Azure.
Culture & Benefits
Part-time student/intern position on Shift 1 in Poland.
Hybrid work model combining on-site work at the assigned hirify.global site with off-site work.
Opportunity to work with academic, industry, cloud, open-source, and customer partners on AI infrastructure.
Work focused on reproducible research, real hardware, clusters, and findings suitable for internal and external publication.
Будьте осторожны
Если работодатель просит войти в их систему, используя iCloud/Google, прислать код/пароль, запустить код/ПО, не делайте этого. Это мошенники. Обязательно жмите «Пожаловаться» или пишите в поддержку. Подробнее в гайде →