What You’ll Learn

• Project Overview: Develop systems software for memory and storage pools that support reinforcement-learning post-training of large language models
• Skills You’ll Learn:
  • Core Linux kernel memory management for virtual machines
  • How to modify the Linux kernel memory/storage management subsystem
  • Contributing to the largest open source project in the world
  • Understand how open-source SW impacts Samsung business
  • Collaborating across team, Samsung, and industry

What You’ll Do

The Samsung Global Ecosystem Team (GOST) is seeking an intern to develop systems software for memory and storage pools that support reinforcement-learning post-training of large language models. The work will examine how rollout generation, reward evaluation, policy updates, and checkpoint recovery use shared resources. You will build a focused prototype using fabric-attached memory and SSD pool, and measure its effect on training progress, memory pressure, and recovery time.

GOST consists of a geographically distributed group of engineers making high impact contributions both in software and leadership to open ecosystems such as the Linux kernel and QEMU. This work supports Samsung’s latest storage and memory technologies.

As an intern, you will contribute to the SW ecosystem R&D efforts targeting Samsung’s latest memory and storage technologies globally through distributed open-source projects. The goal of the internship is to deliver a working memory-to-storage data path, a live RL post-training demonstration, and a measured explanation of when pooling improves training efficiency or recoverability.

Location: Daily onsite presence at our San Jose headquarters in alignment with our Flexible Work policy

Reports to: Principal Engineer

• Responsibilities

  • Profile the RL post-training loop to identify memory-capacity limits, data-transfer overhead, storage stalls, and GPU idle time.
  • Develop one runtime feature that coordinates bounded memory staging, asynchronous storage writes, prefetching, or capacity allocation across workers.
  • Integrate the feature with an existing RL framework using its training and checkpoint interfaces.
  • Preserve snapshot consistency, policy-version provenance, buffer ownership, and durable checkpoint publication during asynchronous operations.
  • Compare memory-only, storage-only, and combined pool configurations using training throughput, GPU-hours, local-memory usage, and restart performance.
  • Deliver correctness tests, reproducible experiments, design documentation, and code suitable for review and potential open-source contribution.
  • Find the resource management issue on post-RL workload and resolve the issue in host OS.
  • Complete other responsibilities as assigned.
  • Complete other responsibilities as assigned.

What You Bring

• Pursuing Masters, or PhD in Computer Science
• Must have at least 1 academic quarter/semester remaining
• Programming experience in C, C++, or Rust, with Python or shell scripting for experiments and automation.
• Understanding of operating systems, memory management, concurrency, and computer architecture.
• Experience developing and debugging on Linux, using Git, and communicating technical results—coursework and research projects count.
• CXL, NUMA, Device-DAX, mmap, shared-memory programming, or memory allocators, preferred
• GPU programming, CUDA, DMA, or profiling memory and data-transfer performance, preferred
• Distributed systems, cache management, vllm, sglang, LLM inference, or open-source development, preferred
• You’re inclusive, adapting your style to the situation and diverse global norms of our people.
• An avid learner, you approach challenges with curiosity and resilience, seeking data to help build understanding.
• You’re collaborative, building relationships, humbly offering support and openly welcoming approaches.
• Innovative and creative, you proactively explore new ideas and adapt quickly to change.