Research Scientist

present of AI; Palo Alto (CA); On-site; USD 150,000 - 230,000; Full time

Job summary

Simular in Palo Alto is seeking a researcher to lead original work on AI agents that see and act on desktop applications, covering planning, reinforcement learning, and multimodal reasoning. You will design fair evaluations and robust tests to measure agent behavior.

From data collection and benchmark construction to training models and analyzing results, you will produce prototypes for engineers to productize.

Qualifications

• PhD or equivalent research experience.
• First-author papers at major conferences.
• Strong PyTorch or JAX skills.
• Ability to design rigorous agent evaluations.

Responsibilities

• Lead original research on AI agents that use computers, including planning, RL, and multimodal reasoning.
• Collect data, build benchmarks, train models, analyze results, and hand prototypes to engineers.

Skills

PyTorch

JAX

First-author papers

Education

PhD or equivalent research experience

Job description

This role is for someone who can lead original research on AI agents that use computers, covering planning, reinforcement learning, multimodal reasoning and ways to judge agent behavior. The work runs from collecting data and building benchmarks to training models and analyzing results, then handing prototypes to engineers for the product. A strong candidate has a PhD or equal research experience, first author papers at major conferences, and solid PyTorch or JAX skills.

How it fits the AI era

The job sits at the center of the shift toward AI that does tasks on a screen instead of only answering questions. AI tools speed up coding, experiment setup and data analysis, but choosing which problems matter, designing fair evaluations and judging whether an agent can be trusted stay with people. Careful thinking about safety and reliability becomes more valuable as agents take on real work.

How it builds AI experience

Someone in this role would work directly on training and testing models that see a screen and act on desktop applications, which is hands-on experience at the frontier of agent research. To build toward it, try fine tuning a vision language model on a small computer use task, or reproduce results on a public agent benchmark and write up what you found.

Worth showing

• LLM or VLM fine tuning
• Reinforcement learning for agents
• PyTorch or JAX experiments
• Benchmark and dataset building

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 Simular

Autonomous computer use platform developing research driven artificial intelligence agents to operate desktop applications and manage repetitive knowledge tasks