Post-Training Research Scientist

Two Sigma; New York (NY); On-site; USD 180,000 - 240,000; Full time

Job summary

Two Sigma seeks a researcher-focused engineer to lead post-training efforts for language models applied to financial time series and quantitative reasoning. You will design RLHF, DPO, and related methods at scale, managing substantial compute budgets and building data/reward infrastructures on financial datasets.

You will collaborate with quant researchers to align post-training work with finance applications, driving a strong research agenda and ensuring smooth production integration.

Qualifications

• Shipped post-training systems in production (RLHF, DPO, RLAIF) with alignment goals.
• Strong understanding of post-training methods and alignment research.
• Hands-on experience with PyTorch/JAX and distributed frameworks.

Responsibilities

• Lead post-training efforts for LLMs applied to financial time series and quantitative reasoning.
• Design and execute RLHF, DPO, and related alignment methods at scale, including deployment of substantial compute budgets.
• Build infrastructure for preference data collection, reward modeling, and policy optimization on financial datasets
• Drive research agenda connecting post-training methods to quantitative finance applications
• Collaborate with quant researchers to define task distributions and evaluation frameworks
• Unblock production systems dependent on post-training capabilities

Skills

PyTorch/JAX

Distributed training

Experiment design

Budgeting & compute management

Education

BS or MS in STEM

Tools

DeepSpeed

FSDP

Multi-node GPU clusters

Job description

Responsibilities

• Lead post-training efforts for LLMs applied to financial time series and quantitative reasoning
• Design and execute RLHF, DPO, and related alignment methods at scale, including deployment of substantial compute budgets (O($100mm))
• Build infrastructure for preference data collection, reward modeling, and policy optimization on financial datasets
• Drive research agenda connecting post-training methods to quantitative finance applications
• Collaborate with quant researchers to define task distributions and evaluation frameworks
• Unblock production systems dependent on post-training capabilities

Qualifications

• BS or equivalent work experience in Science, Technology, Engineering or Math (an MS is a plus).
• Minimum 1 year of experience required; 1-10 years of experience preferred (ideally 1-5 years) at a frontier AI lab (OpenAI, Anthropic, DeepMind, Meta FAIR, or equivalent)
• Shipped post-training systems in production: RLHF, DPO, RLAIF, or related methods
• Deep understanding of distributed training infrastructure: multi-node GPU clusters, training stability, checkpointing
• Track record managing large-scale compute: budgeting, experiment design, ablations
• Publications or demonstrated expertise in alignment, preference learning, or reward modeling
• Hands-on implementation skills: PyTorch/JAX, distributed frameworks (DeepSpeed, FSDP, etc.)