AI Prompt Engineer Jobs in India (Remote, Full-Time) — Rex.zone

Design, test, and optimize prompts and evaluation workflows that improve LLM performance in real production pipelines. You will collaborate with engineers and researchers to build prompt libraries, run A/B tests, define rubrics, and improve reliability, groundedness, and safety across assistants, copilots, agents, and RAG systems.

Key Responsibilities

• Own prompt lifecycle management: writing, debugging, versioning, and prompt library maintenance
• Design prompt evaluation plans using rubrics for accuracy, completeness, instruction following, tone, and policy compliance
• Create RLHF-style preference/ranking data with clear rationales to support instruction tuning workflows
• Translate product requirements into system prompts, developer prompts, and user prompt patterns
• Improve grounding and reduce hallucinations using RAG strategies and citation/traceability constraints
• Create annotation guidelines and support QA evaluation (edge cases, escalation paths, consistency checks)
• Monitor quality metrics (pass rate,



inter-annotator agreement, regression failures) across releases
• Support safety alignment: red-teaming prompts, jailbreak resistance, and sensitive data leakage prevention

Required Qualifications

• 3+ years in software engineering, ML/NLP workflows, or applied LLM prompt engineering
• Robust structured writing for precise instructions and multi-step constraints
• Experience with LLM evaluation: rubrics, pairwise ranking, golden sets, adversarial testing
• Familiarity with RLHF concepts and human feedback loops
• Working knowledge of Python and experimentation tooling (notebooks, scripts, JSON, APIs)

Preferred Qualifications

• Experience with RAG pipelines, vector search, embeddings evaluation, and grounded answer validation
• Knowledge of content safety labeling, policy taxonomy, and prompt-based guardrails
• Experience with LLM observability/eval tooling (regression suites, eval ha

📌 AI Prompt Engineer (India)
🏢 Rex.zone
📍 India