Role Overview

If you are frustrated by the current boundaries of AI and eager to translate equations from textbooks into billion-parameter models or robots that interact seamlessly with the physical world—join us.

We are looking for passionate, hands-on, and intellectually curious students (open to Bachelor’s, Master’s, and Ph.D. candidates) for our Full-Cycle Internship Program. We don't care about your seniority or academic pedigree; we care about your clean code, your engineering intuition, and your hacker spirit.

You won't be fetching coffee or doing dry data entry here. You will be a core contributor to the exploration of AGI and Embodied AI.

💡 Choose Your Research & Interest Group:

1. Large Language Models & Multimodal Group

• Pre-training & Architecture: Assist in researching next-generation model architectures; participate in optimizing the scaling stability and compute efficiency of massive distributed clusters.
• Post-training & Alignment: Replicate, benchmark, and improve cutting-edge SFT, RLHF, and RLAIF (PPO, DPO) algorithms. Explore how to unlock model capabilities in complex reasoning and "slow-thinking."
• Data Intelligence: Build automated, intelligent synthetic data generation and curation pipelines to feed the next generation of foundational models.
• Adversarial Evaluation: Develop automated evaluation frameworks and Agent systems to precisely benchmark model limitations, hallucinations, and safety boundaries.

2. Diffusion & World Models Group

• Physical World Modeling: Explore physical-law-grounded World Models and tackle core algorithmic bottlenecks in high-consistency, long-horizon video and scene generation.

3. Agentic AI Group

• Complex Agent Systems: Help build production-grade agent frameworks that leverage multimodal perception, long-term memory, autonomous tool selection, and multi-agent collaboration.

4. Embodied AI & Robotics Group

• VLA Foundations: Participate in training Vision-Language-Action (VLA) models to bridge digital intelligence with physical-world decision-making.
• Dexterous Manipulation: Assist in designing RL-based control algorithms for high-DoF (degree of freedom) dexterous hands and tactile-sensing integration to solve compliant, fine-grained robotic manipulation tasks.

🎯 Who We Are Looking For

• Target Group: Current undergraduate or graduate students graduating in 2027 or later (open to early-stage students with high potential).
• Solid Engineering Base: Highly proficient in PyTorch or JAX with strong debugging and code execution skills. Adept at using AI-assisted coding tools (e.g., Cursor, open-source agent frameworks) to maximize research velocity.
• We Value At Least One of These Traits:

1. Paper Replicability & Hacking: The ability to independently digest the latest ArXiv/top-tier AI papers and rapidly reproduce their core results in a clean, reproducible codebase.
2. Competition / Open-Source Enlistee: Proven track records in Kaggle, top-tier AI/Robotics hackathons, or active contributions to well-known open-source AI projects.
3. Strong Theoretical Instincts: A solid mathematical foundation in machine learning, reinforcement learning, 3D computer vision, or robot dynamics.

🌟 Internship Perks & Fast-Tracks

• High-Priority Compute Access: Interns receive dedicated, high-priority GPU cluster allocations. No endless waiting in lines—run your experiments and validate your ideas the same day.
• Dual Track Mentorship (Academic & Industry): 1-on-1 guidance from top industry scientists and conference reviewers. We actively support turning your internship breakthroughs into top-tier workshop/conference papers (with you as first author) or shipping them directly to production.
• Flexible & Accommodating: We offer flexible long-term internship options (minimum 3 months, 6+ months preferred). For certain research-heavy tracks, we support hybrid or remote working arrangements so you can balance your university studies.
• Return Offer Fast-Track: Exceptional interns will bypass standard campus recruiting pipelines and receive immediate fast-tracked Return Offers (including elite SSP/Top Talent packages).