I’m 19, not some hardcore telecom engineer, and I just built a fully custom low-latency AI voice receptionist for a client that’s nearly 90x cheaper than relying on expensive voice AI wrappers like Vapi. Honestly… did we just expose how overpriced some of these platforms are? The stack:• LiveKit• Deepgram Nova-3• Gemini 2.5 Flash• Cartesia Sonic-2• Modal• n8n• Vobiz SIP telephony This wasn’t just another AI chatbot. The system:• answers real phone calls• handles multilingual conversations• interrupts naturally• captures leads automatically• extracts emails from noisy calls• triggers backend automations• runs with extremely low latency One of the funniest problems:The AI once heard:“[email protected]” as:“[email protected]” Ended up building custom spoken-email normalization logic just to make email capture reliable over phone calls. And honestly, n8n was one of the smartest decisions in the stack. Because voice AI systems don’t stop at conversations.You eventually need:• CRM integrations• lead routing• webhooks• follow-ups• notifications• scalable workflows Feels like we’re entering a phase where more builders realize they can create custom voice AI infrastructure with better control, lower latency, and dramatically lower costs instead of depending entirely on expensive plug-and-play platforms. Curious:Do you think custom voice AI stacks will eventually replace most voice AI SaaS platforms?

Build AI voice agents instantly using ChatGPT's new native Retell AI integration. This tutorial shows you how to create voice agents through ChatGPT's interface without writing code or dealing with complex APIs. Check it out 👇