What problem does it solve? Building a low-latency inbound voice call agent requires coordinating speech-to-text, a streaming LLM, and text-to-speech over a single WebSocket pipeline, and small misconfigurations cause silent failures or high latency. This Skill provides the architecture, constraints, and debugging checklist for the Marcela voice agent. ## Core Features & Use Cases - Streaming Voice Pipeline: Connects Twilio ConversationRelay, Deepgram nova-3 STT, Gemini 2.5 Flash streaming, and ElevenLabs Flash 2.5 TTS over one Starlette server port. - Call Screening Logic: Implements layered transfer rules that qualify callers, block spam and recruiters, and route priority contacts to Ricardo. - Multilingual Support: Auto-detects English, Portuguese, Spanish, and Italian with dedicated ElevenLabs voices per language. - Use Case: A developer deploying the agent to Railway hits a form-parsing error; the Skill points to the missing python-multipart dependency and the single-port constraint immediately. ## Quick Start Ask the assistant to help debug or extend the Marcela voice agent, for example by checking why the Twilio webhook is not reaching the WebSocket endpoint.
