Research Intern (Frontier AI Systems) - 2027 Start (PhD)

Job description

About the Team We are a lean architect & research team responsible for defining the next generation of AI infrastructure at Bytedance. AI is a fast-evolving horizon — pretraining, RL, and agentic workloads each reshape the requirements faster than traditional cloud abstractions can absorb — and our team is built to keep pace rather than simply react. We approach the problem as an end-to-end AI factory: a tightly coupled production system spanning data, applications, software infrastructure, chips, energy, and the broader supply chain. In this role, you will work at the intersection of large-scale systems, AI, emerging hardware, and the cognitive foundations of intelligent agents — including next-generation AI memory systems informed by cognitive science and psychology — designing scalable architectures and driving innovations across the full AI factory stack. This internship is intended for PhD students who want to work on technically deep, open-ended problems with real systems relevance. We value people who are self-directed, comfortable with ambiguity, able to move between abstraction and implementation, and excited to turn ideas into working systems, measurements, and technical direction. We are looking for talented individuals to join us for an internship. PhD internships at Our Company provide students with the opportunity to actively contribute to our products and research, as well as to the organization's future plans and emerging technologies. Our dynamic internship experience blends hands-on learning, enriching community-building and professional development events, and collaboration with industry experts. Applications will be reviewed on a rolling basis, so we encourage you to apply early. Please clearly state your availability in your resume (Start date, End date). Responsibilities: - Conduct research and build prototypes for next-generation AI systems spanning large-scale training, post-training, reinforcement learning, inference, retrieval, and agentic workloads. - Study bottlenecks and design trade-offs across the stack, including communication, scheduling, storage, observability, runtime efficiency, memory behavior, and serving infrastructure. - Develop benchmarking and evaluation methodologies that capture not only raw performance, but also system efficiency, scalability, robustness, and cost under realistic AI workloads. - Explore new system designs in areas such as distributed training and RL systems, inference optimization, retrieval and memory architectures, agent infrastructure, and heterogeneous hardware-software co-design. - Work closely with researchers and engineers to turn ideas into prototype systems, empirical insights, and technical proposals that can influence future infrastructure directions. - Communicate results through clear technical writing, experimental analysis, and collaborative discussion. Example Research Directions You may work on one or more of the following areas, depending on background and project fit: - Distributed Training, Post-Training, and RL Systems Study the systems challenges behind large-scale model development pipelines, including communication bottlenecks, cluster efficiency, scheduling, checkpointing, rollout infrastructure, and resource orchestration for RL and post-training workloads. - Inference and Serving Systems Design systems for low-latency, high-throughput, and cost-efficient serving of large models, including KV cache management, batching, runtime optimization, memory efficiency, elastic scaling, and accelerator-aware serving architectures. - Retrieval, Memory, and Long-Horizon Agent Infrastructure Explore infrastructure support for retrieval-augmented systems, AI memory architectures, and long-horizon agents, including vector indexing, memory organization, state management, and coordination across multi-stage agent workflows. - Storage, Networking, and Observability for AI Workloads Develop system designs for high-performance storage, cluster networking, observability, fault localization, and root cause analysis in large-scale AI environments where training and serving workloads operate under tight performance constraints. - AI for Infrastructure Investigate how learning-based methods and AI agents can improve infrastructure diagnosis, tuning, orchestration, scheduling, and system optimization for large-scale AI workloads. Minimum Qualifications - Currently pursuing a PhD in Computer Science, Computer Engineering, Electrical Engineering, Mathematics, Statistics, or a related technical field. - Strong background in one or more of the following areas: ML systems, distributed systems, large-scale training, inference systems, reinforcement learning systems, storage systems, networking, or infrastructure engineering. - Strong systems intuition and analytical ability, with the ability to reason across model behavior, runtime performance, and system-level trade-offs. - Ability to work independently in ambiguous, fast-evolving technical spaces and make progress with limited guidance. - Strong written and verbal communication skills, and the ability to collaborate effectively with both research and engineering partners. - Genuine interest in the future of AI systems and infrastructure. Preferred Qualifications - Experience with large-scale model training, post-training, or reinforcement learning systems. - Experience with inference or serving systems, including distributed inference, KV cache optimization, batching, runtime optimization, or accelerator-aware system design. - Familiarity with retrieval systems, vector indexing, AI memory systems, long-context systems, or infrastructure for agentic workloads. - Familiarity with large-scale infrastructure concepts such as RDMA, NCCL, cluster scheduling, storage acceleration, heterogeneous compute, or HPC-style workloads. - Strong research track record, such as publications in machine learning, systems, architecture, or related venues. - Experience building systems prototypes, resea