The SymptomWise framework separates language understanding from diagnostic reasoning to eliminate hallucinations in medical AI. It maps free-text inputs to validated symptoms and applies deterministic codex-driven inference over a finite hypothesis space. This architecture restricts large language models to extraction tasks only. Practitioners gain a traceable, reliable path for generating ranked differential diagnoses.