AI Safety Data Annotator – Agentic Trajectory Author

Role Overview
We are seeking Data Annotators and Trajectory Authors to support an AI safety training project focused on
stateful agentic data collection. The primary responsibility is to write long, realistic AI agent conversations in
which a policy violation is hidden deep inside a multi-turn session — not visible from any single message. This
role involves authoring both the violation session and a paired clean “benign twin” version, marking the key
turns, and ensuring internal consistency throughout. This is NOT standard content moderation or single-message
labelling. It is closer quality engineering, and it requires strong writing ability, logical reasoning, and attention to
detail across very long sessions.
Requirements
Bachelor’s degree in any discipline.
Strong English writing ability — can write realistic, multi-turn dialogue that sounds like a real customer
support or enterprise workflow.
Basic understanding of how AI tool calls work — structured function calls, argument names, error
outputs, retries.
Attention to detail
Good communication and documentation skills.
Preferred Qualifications
Candidates from the following backgrounds are best suited for this role:
English — strong writing, grammar, and understanding of how language works in context.
Preferred streams: Computer Science / IT / B.Sc. (Computer Science) / BCA — familiarity with AI, APIs,
tool calls, function outputs, and structured data.
Freshers with strong writing ability and logical reasoning skills, and good working knowledge of Python
or Java, are eligible and will be considered.
Exposure to JSONL or structured data formats is an advantage.
Familiarity with OpenAI function calling, or similar AI platforms is a bonus.
Skills: english,it,java,communication,python,computer science

Work arrangement
Yes