AI research scientist

present of AI; San Francisco (CA); Sur place; USD 180 000 - 240 000; Plein temps

Résumé du poste

Writer is seeking a researcher who has trained large language models at scale and can own a project from concept through training, evaluation and release. The ideal candidate writes PyTorch or JAX code and has published at top ML venues.

They should have real experience with agents that plan, use tools, and recover from mistakes in production settings. The role focuses on advancing how models reason, plan and act across multi-step tasks within enterprise data environments, with strong emphasis

Qualifications

• Experience training large language models at scale.
• Strong knowledge of post-training methods.
• Experience with enterprise agents that plan, use tools, and recover from mistakes.
• Publications in top ML venues preferred.

Responsabilités

• Own a project from idea to training, testing and release.
• Experiment with agents that plan, use tools and recover from mistakes.
• Publish findings in top ML venues.

Connaissances

LLM Training

Post-training methods

Agent planning & tools

Outils

PyTorch

JAX

Description du poste

This role asks for a researcher who has trained large language models at scale and can own a project from first idea through training, testing and release. A strong candidate knows post-training methods well, writes their own PyTorch or JAX code, and has published at top venues. They also need real experience with agents that plan, use tools and recover from mistakes.

How it fits the AI era

The work is the AI itself: improving how models reason, plan and act across multi-step business tasks. Automated tools like model-based judges and synthetic data speed up experiments and data building. People still choose which questions matter, decide what counts as good performance, and judge whether a result is trustworthy enough to ship.

How it builds AI experience

Someone in this role would gain hands-on experience tuning models for enterprise agents that work with company data, which is the core of what Writer builds. To prepare, a candidate could build a small agent benchmark, run a post-training experiment on an open model, and write up where the agent fails and what fixed it.

Worth showing

• Post-training at scale
• Published ML research
• PyTorch or JAX pipelines
• Synthetic data and LLM judges

This is our own read, written from the employer's posting and the company's Spotlight. The employer's posting is the source of truth for the role.

About Writer

Writer is an enterprise AI platform for building and supervising AI agents grounded in company data, per the company.