AI Researcher

TL;DR

AI Researcher (Agentic AI): Developing novel AI algorithms and techniques for action-taking agentic AI with an accent on reasoning, long-horizon planning, memory, multi-agent coordination and alignment. Focus on prototyping new architectures, running live A/B tests on billions of real trajectories, and fusing research with production-grade consumer experiences.

San Francisco Office, Full time On-site. Top-tier relocation and immigration support.

Company

Stealth team of elite founders and AI researchers from Stanford, OpenAI, and DeepMind building trustworthy, consumer-grade AGI agents. Supported by tier-1 investors who funded first-generation AI giants.

What you will do

• Prototype new architectures in reasoning, long-horizon planning, memory, multi-agent coordination and alignment.
• Run live A/B tests on billions of real trajectories and ship ideas to production within weeks.
• Pair with product engineers to integrate novel algorithms with delightful consumer experiences.
• Establish research rigor, safety metrics, and evaluation pipelines for agent-level capabilities.
• Mentor interns, residents, and mixed-discipline team members to amplify team impact.

Requirements

• PhD or equivalent track record in ML / NLP / RL / systems
• 3+ strong papers or significant open-source impact
• Mastery of PyTorch or JAX
• Experience with large-scale training across distributed GPU clusters
• Strong programming skills and ability to ship production code

Nice to have

• Publications at top-tier ML conferences (NeurIPS, ICML, ICLR, ACL)
• Hacker mindset — debugging HTML selectors and re-deriving state-space math
• Experience with RLHF, alignment, or agent-based systems
• Track record of research shipped to production
• Passion for shipping to real users

Culture & Benefits

• All in, in person — work moves faster face-to-face
• Ship by default — speed and polish coexist
• One band, one sound — radical candor, zero politics
• Competitive company-sponsored medical, dental, and vision insurance

Hiring process

• Send a link or 60-second video of something you built and why it matters, resume or LinkedIn, and two sentences on the hardest problem you've cracked.
• Every exceptional candidate hears back within 48 hours.