Post-Training Applied Researcher

present of AI; San Francisco (CA); On-site; USD 170,000 - 230,000; Full time

Job summary

present of AI in San Francisco seeks a hands-on researcher to advance post-trained language models using reinforcement learning. You will design rewards, build multi-turn tool-using environments, and shepherd a project from dataset to deployed model, with an emphasis on understanding why methods work and how to scale them.

Ideal candidates have experience with RL methods like PPO or policy gradients, and can translate experimental results into production-ready systems.

Job description

The role is hands‑on applied research in post‑training language models, mostly with reinforcement learning, for specific customer tasks. A strong candidate has trained models with methods like GRPO or PPO and understands why they work, not just how to run a recipe. They can design rewards that hold up at scale, build multi‑turn tool-using environments, and take a project from dataset to deployed model.

How it fits the AI era

AI is the core of the work here, since the job is training models that beat larger closed ones on narrow tasks. Tools can speed up data generation and experiment setup, but judgment about what a reward really encourages, why a run is unstable, and what a customer's domain needs stays with the researcher. Spotting when a model is gaming its score is a human skill that matters more as training scales.

How it builds AI experience

The role gives direct practice running post‑training pipelines for fast‑growing AI companies, with results that reach production users. Someone building toward it could fine‑tune an open model with RL on a small domain task, write their own reward and eval harness, and document where the reward failed. Working on an open‑source training library or reproducing a recent distillation or on‑policy paper would also show relevant depth.

Worth showing

• Reward design and hacking diagnosis
• Eval harness building
• Training to production feedback loop

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 Baseten

High‑performance cloud inference platform for deploying, serving, and scaling open‑source and custom artificial intelligence models.