Research Scientist, Agent Reasoning San Francisco

Pazau Agent 3 plans multi-step work: it checks eligibility, calls three systems, asks for approval and then confirms with the customer, all while following a Procedure a CX manager wrote in plain English. We don't train foundation models on customer data, so our research focuses on the layer above them: planning, tool use, policy adherence and knowing when to hand off.

The base salary range for this role is $260,000 – $340,000
. Final offers depend on level and location, and every offer includes equity.

• Research methods that improve Procedure adherence and Action accuracy without customer-data training
• Design reasoning strategies for long, multi-turn conversations where the customer changes their mind halfway through
• Build calibrated confidence signals that decide when the agent acts, asks for approval or escalates to a human
• Work with the Evaluation team to turn research wins into shipped improvements, measured in customer resolution rates

• A PhD or equivalent research experience in machine learning, NLP or a related field
• A track record of research on LLM agents, planning, tool use or reasoning
• Strong engineering skills. You'll prototype in Python and ship alongside production engineers
• Interest in problems where being right 97% of the time is not good enough

• Publications at NeurIPS, ICML, ACL or similar
• Experience with RL from environment feedback or simulation-based training
• Worked on AI safety, calibration or interpretability
• Medical, dental and vision for you and your dependents, with 100% of premiums covered
• Equity in Pazau, with a 10-year exercise window after you leave
• 16 weeks of fully paid parental leave for every parent
• 401(k) with a 4% match
• $2,000 a year for learning, plus conference travel when you're speaking
• Four weeks of paid time off, plus a company-wide week off at the end of December
• Recruiter call (30 min). We walk through the role, compensation and timeline, and answer your questions first.
• Hiring manager conversation (45 min). A conversation about work you're proud of and the problems this team owns.
• Research talk (60 min). You'll present your past work to the AI team, followed by a technical discussion.
• Final interviews (3–4 hours, virtual or on-site). Three or four conversations with the people you'd work with every day, including one cross-functional partner.
• Founder conversation and references. A 30-minute conversation with Amara or Lukas, then two references you choose.

Most candidates go from first call to offer in three to four weeks. You'll hear back from us within three business days after every stage, and every candidate who reaches final interviews gets written feedback.

We reply to every application within five business days.