Web Research Specialist

About the Role

We are developing an evaluation benchmark for advanced AI browsing agents. Your responsibility is to craft research challenges that a cutting-edge AI cannot solve, even when given complete web access and multiple attempts. This position focuses on investigative research rather than subject-matter expertise or content writing. You will begin with a verifiable fact and reverse-engineer it into a question that makes it extremely difficult to find, supported by a thorough and auditable evidence trail.

Deliverables

• Create natural-language research questions with concise, stable, and objectively verifiable answers.
• Provide independent clues that can be checked separately, covering various fact types such as dates, people, places, organizations, works, events, records, and quantities, each with specific requirements.
• Produce a validation record documenting the obvious searches performed and their outcomes.

Required Qualifications

• Proven skill in open-web research, including locating primary documents and accessing government and institutional databases, archives, registries, and PDF files.
• Precision in sourcing, citing exact pages, tables, and sections rather than general homepages.
• Ability to research unfamiliar topics independently.
• Native or near-native proficiency in written English.
• Strong aptitude for creating detailed structured documentation; the evidence trail constitutes most of the task.
• Experience with large language model (LLM) evaluation, red-teaming, or building benchmarks.
• Background in one or more of the following areas: reference librarianship, archival research, special collections, investigative journalism, professional fact-checking, OSINT, due diligence, KYC, investigative research, patent or prior-art searching, legal discovery searches, genealogy, or developing competitive quizzes and puzzle hunts.

Preferred Qualifications

• Additional experience with LLM evaluation, red-teaming, or benchmark creation.
• Knowledge of JSON and structured data formats.