Prompt Engineering Interview Pack

This folder contains project-specific interview preparation for prompt engineering in MangaAssist.

The focus is not generic LLM trivia. Every question is grounded in the realities of this repo's design: hybrid routing, RAG grounding, structured outputs, guardrails, prompt hardening, latency, cost, and failure recovery.

How To Use This Pack

• Start with basic if you want fluency in prompt architecture and scenario selection
• Use medium to rehearse intent-specific prompting, token budgets, and retrieval-aware prompting
• Use hard and above for failure analysis, guardrail interactions, evaluation design, and production tradeoffs
• Use questions-only for timed practice
• Use with-hints when you want answer-shaping guidance

File Map

Coverage Areas

• template vs API vs RAG vs FM prompt selection
• system prompt design and output contracts
• recommendation prompting and retrieval-aware prompting
• structured-output reliability
• prompt injection resistance and guardrail coordination
• token, latency, and cost optimization
• cases where prompt tuning failed but a system-level workaround succeeded
• evaluation, regression, and rollout strategy

Grounding Rule

All questions assume the MangaAssist design documented in this repository. Where a question pushes beyond what is explicitly implemented, it does so as a design challenge and not as a hidden assumption.