LLM Prompt System Design — Production-Grade Prompt Architecture

I will design a production-grade prompt architecture for your LLM application — covering system prompt design with role, context, constraints, and output format specification, dynamic context injection patterns for user-specific data, few-shot example selection strategy, fallback prompt for edge cases, and a prompt versioning approach enabling A/B testing of prompt improvements. Production LLM applications require a prompt architecture — not a single prompt — because different query types need different instructions, context window limits require dynamic content prioritisation, and ongoing prompt improvement needs version control and testing infrastructure that a single static prompt cannot support.

The architecture covers system prompt design, dynamic context injection templates, few-shot example library, fallback and error prompt design, a prompt versioning document, and an evaluation framework for testing prompt variants against your quality criteria.