About the Role
For all students looking to gain hands-on experience! You'll work directly with senior scientists on research problems at the frontier of LLM reasoning, post-training methodology, and agentic AI — applied to global crypto markets. Your work has a direct path to production systems serving hundreds of millions of users, and where findings warrant it, a clear path to external publication.
You run experiments, implement ideas from recent research, synthesize papers into hypotheses, and work with engineers to understand how research translates to real systems.
This is not a purely literature-review or passive research role. You are expected to think independently and generate insight.
Responsibilities
Contribute to the design and execution of experiments in reasoning model training, post-training alignment, test-time scaling, or systematic model evaluation — with a focus on applications in financial and crypto-native contexts
Review and synthesize recent academic literature at NeurIPS, ICML, ICLR, and ACL — tracking developments in reasoning, alignment, and agentic AI to inform and sharpen ongoing research directions
Implement model variants, training procedures including RLVR-based approaches, and evaluation protocols using PyTorch and the Hugging Face ecosystem
Track and log experiments systematically using Weights & Biases or equivalent — maintaining reproducibility standards throughout
Explore the intersection of LLM reasoning and crypto-native data: on-chain signals, market microstructure, multi-modal market intelligence — identifying research opportunities unique to Binance's position
Collaborate with applied engineering teams to understand how research findings translate into production constraints in a zero-downtime, 24/7 trading environment
Qualifications
Currently pursuing a Master's or PhD in Machine Learning, Computer Science, Mathematics, or related field strongly preferred;
Expected graduation in 2026, 2027, or 2028
Strong Python programming skills and PyTorch proficiency; C++ or Rust exposure a plus. Equally important: demonstrated comfort with vibe coding — using AI-assisted development tools fluidly as part of your research and experimentation workflow
Solid understanding of transformer architectures, large language model pretraining, and the evolution toward reasoning models
