Architecture and Design of a Prototype Multi-Modal Clinical Decision Support System for Integrative Medicine

Integrative medicine practice requires synthesis of multimodal data — clinical history, laboratory results, and physiological monitoring — in a privacy-preserving environment. No existing open-source clinical decision support prototype addresses this combination with local large language model (LLM) inference. To describe the architecture and design rationale of AIM, an open-architecture prototype clinical decision support system for integrative medicine. AIM integrates three validated core modules: (1) a Bayesian differential diagnosis engine with a 443-node knowledge graph; (2) an OCR-based laboratory data parser with reference-range flagging; and (3) a hybrid LLM layer supporting local inference via Ollama/LLaMA 3.2 and cloud inference via DeepSeek R1. An experimental Ze-HRV biomarker module is described separately in Appendix A and is excluded from the core system. A Telegram bot provides multilingual (Russian, English, Kazakh, Georgian) clinician interface. No evaluation data are presented. This report describes system architecture, design decisions, known limitations, and a structured plan for future validation. AIM presents a replicable architecture for privacy-preserving, modular CDSS prototypes in integrative medicine. Independent validation, open-source code release, and replacement of the Telegram interface with a certified platform are identified as mandatory next steps.