AI Medical Voice Agent Development

A real-time, interactive health advisor powered by AI voice conversations. 🔗 Explore it live: ai-medical-voice-agent-fzjj.vercel.app

I built an AI Medical Voice Agent that transforms how users access health advice. Instead of typing symptoms into a chatbot, users can speak naturally and instantly connect with an AI medical professional from 12 specializations for a live voice conversation. Each session is personalized (with unique IDs) and ends with a detailed medical report.

Frontend: ReactJS, Next.js, TypeScript, Tailwind, Shadcn UI, and Aceternity UI for a dynamic, sleek interface.

Voice AI: AssemblyAI (streaming speech-to-text/text-to-speech) + PlayHT (professional voice sets via Vapi.ai).

Medical AI: OpenRouter’s Google Gemini 2.5 Flash model for accurate, context-aware responses.

Database: Drizzle ORM with Neon for secure conversation logging.

Auth/Payments: Clerk for authentication + subscription billing/payment gateway integration.

Democratizing Healthcare: Bridged gaps in accessibility by offering instant, English-language medical guidance via natural speech.

Technical Growth: Mastered real-time voice streaming, AI model fine-tuning, and secure data pipelines. Learned to balance latency and accuracy in live interactions.

User-Centric Design: Tailored UI/UX for accessibility (e.g., clear voice prompts, session summaries) based on early tester feedback.

Latency vs. Quality: Optimized AssemblyAI and Vapi.ai integrations to reduce lag while maintaining voice realism.

Medical Accuracy: Implemented strict validation layers to ensure AI responses align with trusted guidelines (e.g., WHO, Mayo Clinic).

Scalability: Designed the Drizzle ORM database to handle concurrent sessions efficiently—a first for me at this scale.

This project isn’t just tech—it’s a step toward humanizing AI healthcare. By making expert-level advice accessible through natural conversation, I’m helping reduce barriers to early symptom assessment.