Research Scientist - Reinforcement Learning

Optimized, Inc.; San Francisco, Northern (CA, KY); Hybrid; USD 185,000 - 255,000; Full time

Job summary

Optimized, Inc. is building a small, talent-dense team in San Francisco to deploy AI agents into critical supply chains. You will own RL and post-training, shaping reward models, training loops, and evaluations that translate model capability into reliable long-horizon decisions.

This role is on-site 5 days a week in San Francisco, with visa transfers supported. Compensation ranges from $185,000 to $255,000 plus competitive equity, and you will ground learning in real deployment data and ship it

Qualifications

• Have a PhD or equivalent research experience in RL, ML, or a related field.
• Have hands-on experience with reinforcement learning, post-training, or RLHF for LLMs.
• Are comfortable building research prototypes in Python and iterating quickly.
• Understand reward modeling, policy optimization, and evaluation of sequential decision-making.
• Care about real-world impact, and you have driven research through to production.
• Are excited about applying AI to complex, messy, real-world optimization problems.

Responsibilities

• Train agents to act: design and run RL and post-training pipelines that improve how agents plan and execute multi-step work.
• Build reward models: define and train the reward signals that capture what a good supply chain decision looks like.
• Evaluate long-horizon behavior: build evals that measure agent reliability across long, high-stakes workflows.
• Ground learning in reality: use real deployment data and feedback to close the gap between simulation and production.
• Ship research to production: work with engineers to bring training breakthroughs into the live agent platform.

Skills

Reinforcement learning

Python

RLHF

Policy optimization

Sequential decision-making

Education

PhD or equivalent RL/ML research

Tools

PyTorch

TensorFlow

Job description

We're building a small, talent-dense team to deploy AI agents into the most critical supply chains in the world.

You'll own RL and post-training: the reward models, training loops, and evaluations that turn raw model capability into reliable long-horizon decision-making. You'll ground that work in real deployment data and ship it into production.

This role is on-site 5 days a week in San Francisco. Visa transfers (OPT, H-1B) are supported. Compensation range is $185,000 - $255,000 + competitive equity.

What you'll do

• Train agents to act: You'll design and run RL and post-training pipelines that improve how the agents plan and execute multi-step work.
• Build reward models: You'll define and train the reward signals that capture what a good supply chain decision actually looks like.
• Evaluate long-horizon behavior: You'll build evals that measure agent reliability across long, high-stakes workflows, not just single turns.
• Ground learning in reality: You'll use real deployment data and feedback to close the gap between simulation and production.
• Ship research to production: You'll work with engineers to bring training breakthroughs into the live agent platform.

What you'll bring

• Have a PhD or equivalent research experience in RL, ML, or a related field
• Have hands-on experience with reinforcement learning, post-training, or RLHF for LLMs
• Are comfortable building research prototypes in Python and iterating quickly
• Understand reward modeling, policy optimization, and evaluation of sequential decision-making
• Care about real-world impact, and you have driven research through to production
• Are excited about applying AI to complex, messy, real-world optimization problems