Creating complex technical diagrams using conventional tools such as MS Visio, Draw.io, or Lucidchart demands significant manual effort, domain-specific syntax knowledge, and considerable time investment. These limitations impede productivity and widen the gap between technical intent and visual representation. This paper presents Figmatic, an AI-powered multimodal diagram genera- tion platform that translates natural language prompts directly into professional-quality diagrams by leveraging Google’s Gemini large language model to produce structured Mermaid-based diagram code. The system eliminates the need for manual syntax authoring by offering real-time, conversational diagram creation and iterative refinement. Key features include intelligent diagram-type recommen- dation, a live Mermaid code editor with instant visual preview, auto-save revision history, and multi-format export (PNG, JPG, JSON). A repository analysis module further extends the platform by generating architecture diagrams directly from GitHub repositories. Built on a modern Next.js (App Router) stack with TypeScript, Genkit orchestration, and Tailwind CSS, Figmatic follows an Agile develop- ment methodology comprising five structured sprints. Comprehensive testing across unit, integration, system, functional, and usability dimensions validates its reliability and correctness. The system successfully generates accurate, standards-compliant diagrams for UML, ER, flowchart, sequence, and network diagram families, demonstrating that AI-driven automation can substantially reduce the friction of technical documentation and foster inclusive, collaborative visual communication.

Keywords— AI Diagram Generation; Mermaid.js; Large Language Models; Natural Language Processing; Multimodal Interaction; Next.js; Gemini; Software Visualization