Insurance Domain Expert

What you would do

• Review insurance knowledge tasks and AI model outputs, identifying missing behaviors, flawed assumptions, and misapplied policy language
• Write high-quality instruction specifications and produce golden solutions showing how experts solve insurance and actuarial problems
• Design challenging evaluation sets and benchmarks that reflect real-world insurance work and measure meaningful model improvement
• Work with researchers to translate domain judgment into explicit criteria and keep standards consistent across different insurance specialties
• Provide evidence-based feedback explaining why model outputs do or do not meet professional standards

Who they want

• Credentials: Actuarial credential (FSA, ASA, FCAS, ACAS) or valid license for adjusting, underwriting, or broking
• 4+ years of substantive insurance experience at carriers, reinsurers, brokers, adjusting firms, or regulators
• Senior-level progression (Actuary, AVP, Director, Claims Manager, Head of Underwriting) with genuine ownership of business lines
• Specialization in at least one insurance domain: life-annuity, group benefits, property-casualty, health, reinsurance, pricing, or underwriting
• Hands-on LLM use in professional work; Bay Area location with ability to work on-site multiple days weekly

Main skills

What the interview asks about

1. 1.Spotting actuarial errors under plausibility

   Models produce sentences that sound expert but rest on false assumptions. You must evaluate underlying logic, not just surface fluency.

   For example: “An AI proposes a loss reserve using a 10-year tail for a workers comp line typically running 3-year development. How do you assess whether this represents domain misunderstanding?”

2. 2.Defining golden-standard insurance work

   Researchers need explicit examples of correct reasoning, not just abstract rules. You must show the actual work as practitioners do it.

   For example: “Write an instruction spec for evaluating an underwriting decision on a commercial property application. What makes a response genuinely expert versus plausible-sounding?”

3. 3.Translating implicit judgment

   Insurance experts build intuition over years. Turning that into teachable criteria requires clarity and precision for AI training.

   For example: “You know instantly whether a claims adjustment is reasonable. Describe the actual reasoning steps an AI model would need to follow to replicate that judgment.”

4. 4.Designing benchmarks that matter

   Easy benchmarks flatter models; realistic ones reveal gaps. Your benchmarks must separate genuine expertise from pattern matching.

   For example: “Design a benchmark set for pricing competency that would distinguish a model that memorized pricing tables from one that understands actuarial principles.”

5. 5.Cross-domain calibration rigor

   Different insurance lines have different standards. You maintain consistency without flattening legitimate specialty-specific differences.

   For example: “A claims manager and a reinsurance specialist disagree on whether a model response meets standards. How do you resolve this and document the resolution?”

A task you may get

Write a specification for an insurance knowledge task including example inputs, golden solution with reasoning, and rubric for evaluating alternative responses.

How to prepare

• Prepare 3-5 concrete examples of insurance work from your recent practice, including the explicit reasoning steps
• Reflect on mistakes or edge cases you've encountered and why they trip up non-experts
• Review 5-10 LLM responses to insurance questions and practice scoring them for accuracy and reasoning quality
• Document the unwritten rules and judgment calls in your insurance specialty that formal criteria might miss

The facts

Pay

$60–100/hr

Hours

Full time, 40 hours a week

Where

Hybrid · Bay Area, CA

Open to

USA

Field

Finance

Posted

8/19/2026

Places left

10

We wrote this page from the public Mercor listing. It may be out of date, so read the full posting before you apply.