Much of the knowledge that keeps a small business running lives in people’s heads: who handles a task, what happens when something goes wrong, and which exceptions never made it into a manual. TacitFlow explores how conversational AI can help people describe that knowledge and turn it into an editable business-process diagram.
Powered by Google Gemini, the prototype conducts an interview-style dialogue to elicit process details and generates Business Process Model and Notation (BPMN) 2.0 diagrams. Practitioners can inspect the live diagram, clarify their account, and refine the model through conversation or the visual editor.
Features
- AI-Powered Generation — Describe processes in plain English and get instant BPMN diagrams
- Conversational Interface — Iteratively refine diagrams through chat interactions
- Interactive Editor — Visual BPMN diagram editor with real-time updates using bpmn-js
- Process Analysis — Suggestions for AS-IS to TO-BE process improvements
Technology Stack
- Frontend: Gradio web interface integrated with bpmn-js editor
- AI Engine: Google Gemini for natural language processing
- Backend: Python with BPMN 2.0 XML generation
Research context
The 2025 paper “Documenting SME Processes with Conversational AI: From Tacit Knowledge to BPMN” describes a proof-of-concept evaluation in an equipment-maintenance scenario. The assistant produced an AS-IS model, annotated issues, and generated an improved TO-BE variant in about 12 minutes. This initial evaluation explored whether a conversational workflow could make process documentation more accessible to practitioners in small and medium-sized enterprises.