Internship JD

The General Artificial Intelligence (GeneralAI) Group at MSRA is dedicated to fundamental research to advance the frontier of AI and build next-generation AI models and systems. Our team has a proven track record of delivering high-impact, open-source research, including pioneering work on UniLM, InfoXLM, XLM-E, MiniLM(-2), (m)E5, Layout(X)LM(-3), WavLM, BEiT(-3), Kosmos(-2), VALL-E, DeepNet, LongNet, MiniLLM / On-Policy Distillation, (Gated) RetNet, YOCO / Decoder-Decoder Architecture, 1-bit LLMs / BitNet (b1.58 | a4.8 | b1.58 2B4T | v2 | BitDistill | Sparse-BitNet | bitnet.cpp), Q-Sparse / Fully Sparsely-Activated LLMs, MH-MoE / 1-bit MoE, Differential Transformer, LatentLM / Multimodal Latent Language Modeling, RPT / Reinforcement Pre-Training, VibeVoice (TTS | Realtime | ASR), TPT / Thinking Augmented Pre-Training, Agentic Organization / Asynchronous Thinking, Generative Adversarial Distillation / Black-Box On-Policy Distillation, LLM-in-Sandbox, On-Policy Context Distillation, Online Experiential Learning, among others. Learn more: https://aka.ms/GeneralAI (opens in new tab).

Talent Qualification

We are seeking highly motivated interns to work together on defining the next frontier and paradigm of AI models and systems. We especially welcome candidates with a strong background and an abiding passion for LLMs, MLLMs, Multimodal AI (Speech and Vision-Language models), Agentic AI, and AI systems/infrastructure.

Working Location

Beijing

Internship JD

Modern computing is entering a new phase where systems and AI co-evolve.

Historically, systems innovation has enabled advancements in AI. Today, increasingly capable AI models are making this relationship bidirectional: AI can actively accelerate systems innovation itself.

We explore AI-native systems—a new paradigm that treats AI as a first-class citizen in system design, rather than an external workload. Our focus spans systems that are AI-aware, AI-assisted, and increasingly AI-driven.

Our goal is to fundamentally rethink the systems research lifecycle by:

• Accelerating system design, debugging, and optimization using AI
• Shortening iteration cycles for systems research and development
• Enabling 10× faster exploration and validation of new system ideas
• Delivering real-world impact through accelerated innovation

We take a broad view of systems research, including (but not limited to) systems, networking, and hardware.

We are particularly interested in ideas that challenge the static, GPU-centric, and manually engineered nature of today’s infrastructure, moving toward adaptive, autonomous, and co-designed systems.

Research Topics include, but are not limited to:

• Self-Driving Systems Infrastructure

(e.g., LLM-assisted fault diagnosis, autonomous configuration tuning, AI-driven capacity planning and anomaly detection)

• Agentic AI Systems and Heterogeneous Execution (CPU/GPU Co-Design)

(e.g., speculative tool execution for agent workloads, CPU/GPU co-scheduling, dynamic resource allocation for multi-step agent pipelines)

• AI-Assisted System Design, Debugging, and Verification

(e.g., LLM-powered code review for kernel/driver patches, automated root cause analysis, AI-guided formal verification of distributed protocols or mathematical proofs)

• Next-Generation Architectures for Scalable AI

(e.g., disaggregated memory/compute architectures, new accelerator designs, near-data processing for training pipelines)

• Storage, Communication, and Resource Management for Large-Scale AI Workloads

(e.g., KV cache-aware storage tiering, RDMA/NVLink topology-aware scheduling, checkpoint optimization for trillion-parameter models

What You Will Do

• Collaborate closely with MSRA researchers on frontier problems at the intersection of AI and systems
• Explore new problem formulations and system designs for emerging AI workloads
• Leverage AI to accelerate system development, experimentation, and iteration
• Participate in technical discussions and cross-team collaboration

Talent Qualification

Strong foundation in at least one of the following:

• Systems
• Networking
• Computer architecture
• Machine learning
• Solid systems development skills (e.g., C/C++/Rust/Python), with effective use of modern AI tools
• Strong problem-solving ability, with demonstrated fast learning and self-driven exploration

Preferred

• Interest in cross-disciplinary research (AI + systems)

• Demonstrated excellence in at least one of the following:

• Competitive programming or technical competitions

• High-impact open-source contributions

• Strong academic or industry references

• Experience contributing to large-scale or high-impact projects (e.g., foundation models, system platforms), with the ability to deliver meaningful results within an 8–12 week timeframe

Familiarity with areas such as:

• GPU/accelerator programming, distributed training/inference

• LLM systems (e.g., inference, scheduling, KV cache)

• Networking, storage systems, or resource management

Working Location

Beijing

Internship JD

Our team focused on building next-generation multimodal intelligence agent systems, spanning efficient media representation, AI-native knowledge memories, multimodal reasoning, and human-agent interaction. Our research targets three directions:

1. Human-AI Collaborative Experience: Enable richer human–agent collaboration in co-work, co-research, and co-design via context-adaptive interfaces that infer intent, reduce ambiguity, and co-construct meaning in real time.

2. AI Memory & Knowledge Representation: Develop AI-native memory that compresses media into compact representations so knowledge can accumulate, adapt, and transfer efficiently and securely across agents and interactions.

3. Multimodal Foundation Models: Build unified and efficient multimodal models for coherent, long-horizon agentic reasoning and for semantic, secure, efficient agent-to-agent communication.

Talent Qualification

• Background in one or more: multimodal foundation models, AI agents, HCI, or communication systems.
• Demonstrated research and/or development strength:
  1. Lead or core author of high-impact research papers or technical reports (e.g., highly cited work, awarded paper, etc.,) in Computer Vision, Multimedia, HCI, or Systems.
  2. Built and maintained widely used open-source projects (e.g., high GitHub stars), achieved top rankings or awards in research competitions, and/or contributed to flagship AI systems.
  3. Hands-on experience training large-scale foundation models and/or building agent systems with large number of users.
• Self-motivated fast learner; able to identify key research problems independently and iterate quickly.
• Strong collaboration skills in team-oriented, multidisciplinary research environments.
• Strong human–AI collaboration skills; able to leverage AI tools to assist the full research lifecycle (problem framing, literature review, experimentation, analysis, and writing).

Working Location

Beijing

Internship JD 1: Spatial Intelligence Researcher

Research area: Multi-modality, Embodied AI, World model

Specific Job Description: Participate in the research and development of models to drive technological innovation in the fields of Computer Vision, Embodied AI, Robotics Foundation Model, and World Models. 

Talent Qualification

• Proven experience in one or more of the following areas: 2D/3D Computer Vision, LLM/VLM/VLA, Video Generation, Robotics, Reinforcement Learning, or Generative Models.
• Strong programming skills (e.g., Python, C++) and experience with deep learning frameworks (e.g., PyTorch, TensorFlow).
• Solid understanding of machine learning fundamentals and model optimization.

Preferred

• Experience with large-scale model training, multimodal fusion, or world model learning.
• Publications in top-tier conferences (e.g., CVPR, ICCV, NeurIPS, ICLR, RSS).
• Experience deploying models in real-world systems or products.

Working Location

Beijing 

Internship JD 2: Robotics Hardware Engineer

Research area: Embodied AI, Robotics Foundation Model

Specific Job Description: Participate in the development of cutting-edge Robotics Foundation Model for physical AI. Design, build, and maintain robotic hardware platforms and integrated software stacks. Develop efficient data acquisition pipelines for large-scale model training and validation. Optimize model performance such as efficiency and robustness.

Talent Qualification

• Strong expertise in robotics hardware development, mechatronics, and control systems.
• Proficiency in programming and familiarity with robotics frameworks.
• Experience in system integration, debugging, and performance optimization.

Preferred

• Hands-on experience with robotic platforms (e.g., manipulators, mobile robots, dexterous hands, or teleopration systems).
• Experience supporting data collection for machine learning or robotics model training.
• Experience optimizing AI model’s performance on real robots.

Working Location

Beijing 

Internship JD

Our team develops AI that both learns from the brain and serves the brain, spanning brain-inspired AI, brain–computer interfaces, and AI for understanding and treating neurological disorders. In brain-inspired AI, we study how the brain achieves strong capabilities with minimal energy and translate these principles into more sustainable AI, including brain-like architectures and learning rules as well as brain-inspired embodied AI. In brain–computer interfaces, we advance non-invasive EEG decoding to better characterize brain states and enable practical BCI systems, and address data scarcity and signal noise through foundation-model approaches. In AI for brain health, we apply AI to improve neurological care by enabling more accurate diagnosis from multimodal clinical data, accelerating mechanistic understanding and enabling personalized treatment of brain disorders.

Applicants are expected to collaborate closely with researchers across related areas to investigate high-impact topics. You will contribute to the full research cycle, including data analysis, algorithm design, and experimental evaluation. You will also work with the team to publish in leading AI and interdisciplinary conferences and journals.

Talent Qualification

1. Strong motivation to pursue world-class research through publishing in leading conferences and journals and/or building tools that can meaningfully advance the field. 

2. Proficiency in at least one mainstream programming language (e.g., Python).

Preferred

1. Preferred to be currently enrolled in a Master’s or PhD program in a relevant field (e.g., computer science, neuroscience, etc.), with strong English communication skills (speaking, listening, reading, and writing).

2. Preferred qualifications include publications in top-tier venues and/or research experience in foundation model, multimodal learning, and interdisciplinary AI research.

Working Location

Shanghai

Internship JD

We are looking for passionate, high-potential students to join us in redefining the relationship between AI and society. As we move from static models to autonomous agents, our team focuses on ensuring these systems are safe, aligned, and deeply integrated into human life. You will work on cutting-edge research that bridges the gap between technical breakthroughs and real-world impact.

As a Research Intern, you will drive end-to-end research cycles, from conceptualizing novel algorithms to building scalable systems. You will focus on one or more of the following frontier areas:

• AI Safety & Alignment: Developing robust frameworks to ensure Large Language Models (LLMs) and Agents behave reliably and ethically.
• Next-Generation Evaluation: Moving beyond static benchmarks to dynamic, agentic, and human-centric evaluation methodologies.
• Personalized & Adaptive AI: Researching techniques for AI to understand long-term user preferences while maintaining privacy and safety.
• Social Computing & Simulation: Leveraging multi-agent systems to simulate complex social behaviors and organizational dynamics.
• Agentic AI for Productivity: Building intelligent agents that don’t just “chat,” but execute complex workflows in education and professional environments.

Talent Qualification

• Evidence of high-quality research or equivalent impactful technical projects.
• You must have superior hands-on coding skills and the ability to rapidly prototype complex agentic systems.
• A deep curiosity to understand the underlying mechanics of LLMs and multi-agent interactions.
• Excellent English communication skills for collaborating in a global research environment and writing high-impact papers.

Preferred

• Background or strong interest in Sociology, Psychology, or other social sciences, with a focus on bridging these fields with AI research.
• Experience with frameworks or infrastructure for large-scale model training/fine-tuning.
• The ability to visualize how a research paper transforms into a tool that changes how people learn or work.

Working Location

Beijing