Description
Summary We've built the backbone of an automated outbound-calling voice agent and need an ElevenLabs Conversational AI specialist to take it to production quality. The agent places German-language phone calls to companies to verify whether a specific advertised job position is still open, and returns a structured outcome (open / filled / unclear). A lot is already working — this is a finish-and-polish project, not a build-from-scratch. Already in place: ElevenLabs Conversational AI agent (German) with system prompt, dynamic variables, and structured data-collection outputs Twilio phone number imported into ElevenLabs; outbound calls already work end-to-end Post-call structured extraction (open / closed / unclear + "reached HR?" + transcript) Make.com access and an existing sibling scenario to model the automation on What we need you to do 1. Make the voice sound genuinely natural and human — top priority The current calls sound too "clean" and synthetic. We want it to feel like a real person calling from an office: Background office ambience / room tone (ElevenLabs background_sound) Natural filler words and disfluencies ("hmm", "moment", small pauses) instead of dead silence Human pacing, prosody, and smooth turn-taking; no robotic artifacts or awkward gaps Dial in the right ElevenLabs config for an 8 kHz phone line (voice, model, stability, latency, soft-timeout/filler settings) 2. Robust German conversation handling Switchboard → politely ask to be transferred to HR / Personalabteilung → ask about the one specific role Gracefully handle: ambiguous answers, "Who's calling?", "apply online", voicemail Do not hang up abruptly — wait for the other person's reaction before ending 3. Telephony reliability Retry logic, business-hours call windows, voicemail detection, sensible pacing / rate-limiting 4. Automation & CRM integration (Make.com) Finish the Make.com scenario: trigger outbound calls from a queue, receive the post-call webhook, and write the structured outcome back to Close CRM (custom field + activity) plus an Airtable call log 5. Test & QA pass Run real test calls, tune by ear until it sounds natural, and document the final configuration