The Role and Impact
Intel is building the open software stack that makes large language models run fast on Intel GPUs. As an intern, you will join the engineering team that develops and optimizes LLM inference frameworks. You will work in vLLM and in the GPU kernel libraries beneath it with experienced engineers who will review your code and mentor you. This is a hands-on engineering internship, not a side project. You will write and test kernels, measure where performance goes, and see your changes land in open-source projects used across the AI industry.
Our internships usually last 12-24 months, and you will work with us part-time alongside your university degree. All applicants must be current university students and must remain students for the length of the internship. Candidates must be based in Poland on the internship start date and for its full duration.
Key Responsibilities
- Develop and optimize AI framework software for LLM inference on Intel GPUs, under the guidance of senior engineers, with contributions going upstream to open-source projects such as vLLM.
- Implement, port, and tune compute kernels (e.g., Triton, SYCL) for core transformer building blocks such as attention, sampling, normalization, mixture-of-experts, and quantization.
- Run benchmarks and profile inference workloads to find performance bottlenecks, measure them, and report findings to the team.
- Write and extend unit tests, correctness checks, and performance regression tests so that framework and kernel code stays reliable as upstream moves.
- Investigate and debug failures across the stack, from Python framework code down to kernel behaviour on the hardware.
- Collaborate with framework, compiler, and kernel engineers across global teams, and take part in code review and technical discussion with the open-source community.
We offer a total compensation package that ranks among the best in the industry. The minimum Annual Salary for this job which could be performed in Poland is: 68 880,00 zł. This role will be eligible for our hybrid work model which allows employees to split their time between working on-site at their assigned Intel site and off-site.
Minimum Qualifications
- Currently pursuing a BS or MS degree in Computer Science, Computer Engineering, AI, Mathematics, Physics, or a related field. Student status is required for the whole internship, preferably for the next two years and for at least 12 months.
- Good Python programming skills.
- Understanding of deep learning and neural networks, including the basics of transformer architecture and LLM inference.
- Familiarity with Linux environments from a user perspective.
- Strong analytical thinking and problem-solving skills.
- Eagerness to learn, an open mind, and the ability to work in a team.
- Ability to communicate clearly and concisely in English.
Preferred Qualifications
- Hands-on experience with PyTorch.
- Exposure to GPU programming (Triton, CUDA, SYCL, OpenCL, or similar), for example through coursework or personal projects.
- Working knowledge of C or C++.
- Experience testing and debugging Python code, including unit testing.
- Familiarity with vLLM or another LLM inference serving framework.
- Contributions to open-source projects, or a public portfolio of code (e.g., GitHub).
- Coursework or interest in computer architecture, parallel programming, or numerical methods.
