Expert Voice AI Engineer (Vapi + 11 Labs) - Advanced S2S Sequential Routing & TTS Fallback

Job Description CRITICAL REQUIREMENT To prove you are a human professional and not an AI automation scraping jobs, you MUST start your application cover letter with the exact word: Natillas. If this word is missing at the very beginning of your proposal, you will be automatically rejected without review. Overview We are an established digital agency in Spain looking for a top-tier Conversational AI Voice Engineer to architect and build a high-converting, low-latency B2B Sales Setter Voice Agent. This project requires a Hybrid Speech-to-Speech (S2S) and Audio Playback architecture using Vapi and ElevenLabs. The core setup must be delivered as a modular, easily replicable Blueprint (Template), enabling our internal Junior CTO to duplicate the infrastructure for future agency clients (such as tax advisors, dentists, etc.) by simply swapping knowledge bases and audio assets. The agent must communicate in perfect, natural Spanish (Spain - Castilian). Technical Stack and Requirements Orchestration Platform: Vapi (Advanced Workflows, Custom Nodes, and Intents mapping). Voice Engine: ElevenLabs (Professional Voice Cloning - PVC, and Speech-to-Speech conversion). LLM Provider: OpenAI (GPT-4o / GPT-4o-mini optimized via caching for low latency). STT Engine: Deepgram (Nova-2 optimized for Spanish, custom endpointing and smart formatting enabled). Integration: GoHighLevel CRM via Webhooks/Make. Core Project Scope and Architecture Core Sales Workflow (5 Nodes): Implementation of a linear 5-step qualification script (Introduction, Empathy/Problem Discovery, Diagnosis, Offer Presentation, and Lead Capture). Sequential Objection Handling (State Counters): You must implement conditional logic within Vapi Workflows using session counters. For instance, if a lead triggers the "Price Objection" intent multiple times, the system must sequentially cycle through Answer 1, Answer 2, and Answer 3 (playing distinct pre-recorded S2S audio files for each stage to maintain absolute human realism). Automated TTS Detection and Logging System: When the agent faces an unmapped question and falls back from static audio playback to dynamic LLM text generation (TTS via ElevenLabs), the system must trigger an external webhook via Make/Zapier. This webhook will log the exact question and dynamic response into our database, notifying our team via Slack or WhatsApp to record a new voice FAQ to update the template. Strict Low-Latency Execution: The entire system must run with a node-to-node latency of under 1 second. Proven experience in prompt caching and STT endpoint tuning is mandatory. Scalable Blueprint Structure: The setup must be clean, heavily modular, and designed as a template so our internal Junior CTO can easily duplicate it and swap the variables (such as customer.name or customer.company) and audio files for future clients. Required Deliverables (Milestones) Milestone 1 (Architecture and Prompts): Complete setup of the 5 core conversational nodes in Vapi with clean state transitions. Milestone 2 (FAQ Matrix and Intents): Integration of the sequence-based intent routing with audio playback links for the objection shortcuts. Milestone 3 (GoHighLevel Integration and QA): We will skip this step to first manual-testing the Agent. Working end-to-end automation with Make/CRM and latency optimization certified under 1 second. Milestone 4 (SOPs and Handover Documentation): Comprehensive handover documentation (Standard Operating Procedures) and a series of technical video walkthroughs (Loom) explaining how our Junior CTO can independently duplicate, edit, and launch this template for new clients. Qualifications Proven track record of deploying live production-grade voice agents using Vapi and ElevenLabs. Deep understanding of optimizing STT endpointing, prompt caching, and network payloads to minimize latency. Native or Fluent Spanish speaker (or extensive experience deploying agents tailored to the nuances of the Spanish market). Ability to write clean, maintainable systems and high-quality technical documentation. Obligatory: You must provide proof or case studies of similar complex voice architectures successfully deployed with Vapi. If you have not built sequence-based intent routing before, please do not apply.