AI research scientist at Writer in San Francisco, CA

Job description

• Application
• AI research at WRITER isn't just about publishing papers — it's about building the scientific foundation that powers some of the most ambitious enterprise AI deployments in the world. As an AI research scientist, you'll be at the center of that work. You'll drive a high-impact research agenda focused on large language models, agentic reasoning, and the system-level capabilities that make AI genuinely useful at enterprise scale. This is a rare opportunity to do research that matters twice over — advancing the field and shipping directly into products used by hundreds of thousands of people every day.
• We're at an inflection point. Enterprises are moving from experimenting with AI to deeply embedding it across their operations, and WRITER's models are the engine making that possible. The work you do here — on post-training, planning, multi-step reasoning, and agentic workflows — will directly shape how the next generation of enterprise AI behaves, performs, and scales. You'll have the resources, infrastructure, and cross-functional support to pursue ambitious ideas and bring them to life quickly. This role is hybrid, based out of our San Francisco or New York City hub.
• You'll report to our VP of AI research. Please note: role is open to Senior, Staff, and Sr. Staff applicants
• We are open to hiring for this role at multiple levels, with compensation ranges scaled to match your experience, expertise, and the scope of the role.

Responsibilities

• Lead an independent, high-impact research agenda on large language models and agentic systems, owning projects from early hypothesis through model training, evaluation, and production deployment
• Design and execute large-scale post-training experiments using supervised fine-tuning, reinforcement learning from human feedback (RLHF), RLAIF, DPO, and emerging alignment techniques — with a focus on improving multi-step reasoning, planning, and tool use in enterprise agentic workflows
• Build novel evaluation benchmarks and methodologies that push beyond existing limitations, establishing rigorous measures for how well models perform on complex, real-world enterprise tasks
• Develop scalable data synthesis and curation pipelines that generate the high-quality training signal driving model improvement — including LLM-as-judge frameworks, synthetic data generation, and adversarial dataset construction
• Shape WRITER's model architecture and training roadmap by translating your research insights into concrete improvements to our enterprise-grade LLMs, working hand-in-hand with research engineering and product teams
• Publish and present original research at top-tier venues — NeurIPS, ICLR, ICML, ACL, and others — representing WRITER at the frontier of the field and contributing to the broader scientific community
• Mentor and uplevel fellow researchers and engineers on the team, helping set a high bar for scientific rigor, experimental design, and research culture

Requirements

• 7+ years of hands-on ML research experience, with deep expertise in large language model pre-training and post-training — you've trained models at scale, debugged distributed jobs, and shipped improvements that made a measurable difference
• A Ph.D. in Computer Science, Machine Learning, NLP, or a related field — or equivalent demonstrated research experience with a strong portfolio of independent, published work
• Expert-level knowledge of post-training methods including SFT, RLHF, RLAIF, DPO, GRPO, and related alignment and reasoning techniques, with a track record of applying them to real, production-grade systems
• Strong command of Python and PyTorch (or JAX), with the engineering depth to build and scale training pipelines, evaluation infrastructure, and data synthesis workflows yourself — not just direct others to do it
• A meaningful publication record at competitive ML/AI venues (NeurIPS, ICLR, ICML, ACL, EMNLP, or equivalent), evidencing your ability to originate ideas and execute on a multi-month research agenda independently
• Hands-on experience designing or evaluating agentic systems — models that plan, reason through multi-step tasks, use tools, and recover gracefully from errors — with a nuanced understanding of where they break and how to fix them
• The instincts and orientation that match WRITER's values: you Connect — you collaborate openly across research, engineering, and product and communicate complex ideas with clarity to both technical and non-technical audiences; you Challenge — you ask the hard questions, push back on conventional wisdom, and pursue the research directions others haven't tried yet; you Own — you drive your projects end-to-end with urgency, take accountability for results, and care deeply about the impact your work has on real customers

Skills

Required

• Python
• PyTorch
• JAX
• SFT
• RLHF
• RLAIF
• DPO
• GRPO
• Designing or evaluating agentic systems
• Building and scaling training pipelines
• Evaluation infrastructure
• Data synthesis workflows

Benefits

• Generous PTO, plus company holidays
• Medical, dental, and vision coverage for you and your family
• Paid parental leave for all parents (16 weeks)
• Fertility and family planning support
• Early-detection cancer testing through Galleri
• Flexible spending account and dependent FSA options
• Health savings account for eligible plans with company contribution
• Wellness stipend for gym, massage/chiropractor, personal training, etc.
• Learning and development stipend
• Company-wide off-sites and team off-sites
• Competitive compensation, company stock options and 401k
• WRITER is committed to transparent, market-based compensation practices. Compensation offered will be determined by multiple factors such as role scope and complexity, location, experience, knowledge, and skills. Cash compensation is only one part of WRITER’s competitive Total Rewards package, which can also include equity, an immersive purpose-driven culture, career development, and thoughtfully designed benefits and well-being offerings.