The process began with defining clear constraints around what CHEFED should and should not do. Instead of starting with features, I focused on behavior, tone, and standards. I treated the AI as a product surface that required the same level of UX design as any interface. This included defining when the AI should provide full recipes, when it should explain technique, and when it should refuse a request altogether. Design exploration followed a conversational-first approach. I used AI to rapidly prototype interaction flows, test response pacing, and explore how different tones affected perceived trust and expertise. These explorations informed the final UX and visual design, which intentionally remained minimal and restrained to keep attention on the guidance itself. On the technical side, I designed the AI system as a domain-restricted expert rather than a general chatbot. I implemented prompt-level guardrails that enforced classical technique, ingredient logic, and quality thresholds. Ingredient-based recipe creation was structured around proven culinary reasoning, ensuring that generated recipes remained realistic and professionally sound. AI was also used during development to accelerate front-end implementation, assist with component scaffolding, debugging, and refactoring, and validate accessibility and responsiveness. While AI increased development velocity, all architectural decisions, UX logic, and final outputs were reviewed and refined manually to maintain consistency and quality. Throughout the process, iteration was continuous. The system was repeatedly tested with edge cases, incomplete ingredient lists, and ambiguous user prompts to ensure CHEFED responded with clarity, restraint, and confidence rather than filler or overreach.