### Answer I can share what's publicly documented about Retell AI's architecture, but I don't have access to proprietary backend implementation details. High-Level Process Flow for AI Voice Calls: The core components involved in an AI phone call include: [(1) ]() 1. *Automatic Speech Recognition (ASR) * – Converts what the caller says into accurate text in real time 2. *Large Language Model (LLM) * – Determines the meaning behind the caller's words by detecting and analyzing intent, tone, and urgency 3. Dialog Management & Turn-Taking – Guides the conversation naturally, handling interruptions, confirmations, and multi-step flows 4. Text-to-Speech (TTS) Synthesis – Generates high-quality, human-like responses 5. API & System Integration – Triggers actions like updating a CRM, booking an appointment, or submitting a ticket Telephony Infrastructure: Retell AI uses SIP trunking to connect with telephony systems, enabling compatibility with providers like Twilio, Vonage, Telnyx, Plivo, and RingCentral. [(2) ]() , [(3) ]() Key Technical Characteristics: - Sub-500ms latency for natural conversations [(2) ]() - Concurrent call handling at scale [(3) ]() - Integration with custom LLMs via API However, the specific proprietary implementations, internal algorithms, and detailed system architecture are not publicly documented. If you're looking to build something similar, you'd need to research and integrate ASR providers, LLM services, TTS engines, and telephony infrastructure independently.

​ To learn more, see the sources I considered: [[1] Voice AI](https://www.retellai.com/glossary/voice-ai) [[2] AI Voice Agent Platform for Phone Call Automation - Retell AI](https://www.retellai.com) [[3] Inside Retell AI | The Conversational AI Platform Behind Scalable AI Phone Systems | Retell AI](https://www.retellai.com/blog/inside-retell-ai-conversational-ai-phone-system) ​

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.

@AI Evy I mean to start the conversation do they create a room like LiveKit or it is just sending speech to text and text to speech back

### Answer The available sources don't document whether Retell AI uses a room-based architecture like LiveKit or a different approach for managing real-time voice sessions. What is documented is the **speech processing pipeline**: - *ASR (Automatic Speech Recognition) * converts the caller's speech into text in real time [(1) ]() - *Voice Activity Detection (VAD) * detects when the caller starts and stops speaking - Turn-Taking Endpoints determine when it's the AI's turn to speak vs. listen - TTS (Text-to-Speech) Synthesis converts the AI's response back into speech - Low-latency audio streaming for real-time experiences [(2) ]() The sources describe this as a continuous real-time stream with latency optimization at every step [(1) ]() , but the underlying session management architecture (room-based vs. direct stream) isn't specified in the public documentation. If you need specifics on the internal architecture for building something similar, that level of detail isn't publicly available. You may want to reach out via [Contact Sales]() for more technical discussions.

​ To learn more, see the sources I considered: [[1] Speech Processing](https://www.retellai.com/glossary/speech-processing) [[2] Top 5 Thoughtly Alternatives for AI Voice Agents in 2025 | Retell AI | Retell AI](https://www.retellai.com/blog/thoughtly-ai-voice-agent-alternatives) ​

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.

I am interested in your idea and I can work as a fullstack and AI engineer

I wanna work with you if you need developer