Automation of software development for psychiatry, psychotherapy and medical psychology: methodology for creating technical specifications and code generation using artificial intelligence (vibe-coding)

Context and Background. The development of personalized software for psychiatry, psychotherapy, and medical psychology is complicated by the interdisciplinary barrier between clinicians and developers. The widespread use of large language models (LLM) and intuitive development environments opens up opportunities to automate the creation of specialized solutions, reducing development time from weeks to days. Theoretical Basis. The methodology is based on the spiral software life cycle model (ISO/IEC 12207:2008), which ensures iterative adaptation to the dynamic re-quirements of medical tasks. The integration of LLM into the code generation process is formalized through universal technical specifications (TS). Objective: to develop a methodology for creating deterministic TS for generative AI mod-els, ensuring automated code generation for highly specialized tasks (assessment of comorbid pathology, addiction risks, fatigue as a predictor of neuroticism). Materials and Methods. Free-format technical specifications in Russian with iterative adjustments by experts. Code generation using LLM Qwen2.5-Max (medical terminology support, 131 thou-sand context tokens). Implementation of prototypes in Python 3.13 with the Tkinter library for GUI. Validation of a modular architecture for processing heterogeneous data (questionnaires, audiovisual markers). Results. A functional prototype for predicting medical risks with a multi-window interface and color indication of results was created. 98% of the generated code complied with the technical specifications after two iterations of refinement. Dynamic adaptation of modules (A/B/C) for the tasks of screening for depression, anxiety, and fatigue was implemented. Conclusions. The combination of formalized technical specifications and LLM accelerates the development of medical software, but re-quires interdisciplinary interaction at the requirements verification stage, strict ethical audit (in accordance with GOST R 71657-2024 and Federal Law No. 152), integration with IoT devices (neurovisors, biosensors) for multimodal data anal-ysis. It is recommended to use the methodology for the mass development of personalized tools in the context of a spe-cialist shortage. Key Limitations: dependence on the quality of technical specifications, the inability of AI to offer inno-vative architectural solutions, the need for manual adaptation for legacy technology stacks.