Full-Stack VoIP/AI Engineer

Full-Stack VoIP/AI Engineer – Upgrade Real-Time Voice Prototype to Production SaaS We have a working full-stack prototype for "Voice AI," a multi-tenant platform connecting live phone lines (Twilio Media Streams) directly to the Google Gemini Live API via WebSockets The stack is React 19 (Vite 6 / Tailwind CSS v4) and Node.js (Express / TypeScript) with Supabase (PostgreSQL). We need an expert to fix prototype architectural bugs, implement production-grade SaaS features, and architect a highly scalable, cost-efficient deployment strategy. Audio & Cost Optimization: Streamline the backend binary audio streaming loop to minimize server CPU usage. Live Booking Engine: Map Gemini’s native tool calling to our Supabase Postgres schema so the AI can check real-time availability and manage appointments safely without race conditions. Infrastructure Assessment & Deployment: Propose and execute the ideal architecture (e.g., GCP Cloud Run vs. Self-Hosted Server) ensuring maximum scalability, low latency, and zero-downtime reliability. ### Tech Stack - Node.js (TypeScript), Express, and native WebSockets (`ws`). - Twilio Voice API & Media Streams. - Google `@google/genai` SDK WebSocket Live interfaces. - Supabase / PostgreSQL & Redis caching. For a real-time, WebSocket-heavy app like this, do you recommend deploying via an autoscaling serverless container engine (like GCP Cloud Run) or dedicated, self-hosted infrastructure? Explain briefly how you would ensure connection stickiness, zero call drops during scaling, and minimum server overhead costs.