Chatbot baseado em inteligência artificial para sugestão personalizada de treinos físicos via Whatsapp

Abstract

Sedentary behavior affects approximately 47% of Brazilian adults, worsened by socioeconomic inequalities that limit access to personalized physical activity guidance. This paper presents the development of an artificial intelligence-based chatbot integrated with WhatsApp, capable of suggesting personalized workout plans based on individual user profiles. The solution was built with N8N as a conversational flow orchestrator, two specialized Google Gemini 2.5 Flash agents — a Profile Classifier and a Personal Trainer —, a PostgreSQL 16 database containing 45 academically validated exercises, WAHA as a WhatsApp interface, and Docker for local containerization. Testing was conducted using the black-box method with seven fictional personas, covering three intensity levels, different age groups, distinct physical goals, and varied health conditions. All executions were completed successfully, with an average response time of 19.62 seconds and full respect for all reported contraindications. In three cases the classified intensity level diverged from the expected outcome in a contextually justifiable way, demonstrating that the system integrates multiple profile variables. The results indicate the technical feasibility of democratizing access to physical activity guidance through low-code tools and large language models, in alignment with SDG 3 of the UN 2030 Agenda.