Prompt Architecture Table

Description: The formal prompt engineering framework and constraint matrix designed to govern the LLM’s persona, behavioral guardrails, and dynamic data integration.
What I Did:
Defined System Identity & Persona: Deeply rooted the AI's persona in Fifth Third’s history. Instead of a generic, robotic chatbot, I engineered Jeanie's system instructions to reflect an empathetic, highly secure digital banking assistant with continuity tracing back to her 1977 origins.
Programmed Financial Safety Boundaries: Authoring the strict boundaries for the LLM, I implemented hard instructions stating that the agent is never to provide definitive financial or investment advice, predict market trends, or promise loan eligibility—gracefully pivoting unanswerable compliance queries to live banking staff.
Architected Dynamic Variable Mocking: Configured how the prompt ingests secure real-time metadata (such as authentication status and user first names) to allow personalization without compromising secure backend tables.
Strategic Impact:
By treating the prompt as the actual software design asset, I successfully constrained a non-deterministic LLM to behave with 100% brand consistency, eliminating the risk of rogue AI responses or unauthorized banking promises.