HealStation AI: A Multimodal Medical Consultation System Integrating Large Language Models, Voice Processing, and Location-Based Healthcare Discovery

Access to timely, qualified medical consultation remains inequitably distributed across socioeconomic and geographic boundaries. In India, the physician-to-population ratio falls well below the World Health Organization’s recommended threshold, creating critical delays in triage and specialist referral. This paper presents HealStation AI, an open-source multimodal medical consultation platform engineered to bridge this gap by combining state-of-the-art large language models (LLMs), automatic speech recognition (ASR), text-to-speech synthesis, document vision analysis, and real-time location-based healthcare provider discovery. The system integrates Meta’s Llama-4-Scout-17B multimodal model and Llama-3.3-70B via the Groq inference API, Microsoft Edge-TTS for naturalistic voice synthesis, and OpenStreetMap-powered facility lookup through LocationIQ to deliver end-to-end patient consultation workflows. HealStation AI supports three Indian languages—English, Hindi, and Marathi—processes medical images and PDF laboratory reports, performs rule-based emergency triage using validated clinical scoring instruments (HEART score, BE-FAST, trauma scoring), and recommends appropriate specialists from a taxonomy of twenty clinical domains. A FastAPI backend exposes a well-defined REST API consumed by a React 19 single-page application. Empirical test cases across low, medium, and high urgency symptom profiles demonstrate consistent specialist routing and urgency classification with an end-to-end latency of 6.2 seconds (N=50). The platform is designed for deployment in resource-constrained environments and is structured for extensibility toward telemedicine and Ayushman Bharat Digital Mission (ABDM) integration.