Design AI Voice Companion Automation Pipeline

**PROJECT: AI Voice Companion Pipeline — [login to view URL] + ElevenLabs + Twilio + Airtable + Claude API** --- **OVERVIEW** I am building an AI companion service that makes scheduled outbound phone calls to elderly users. The AI agent (already configured in ElevenLabs Conversational AI) conducts a 7-minute conversation, and the system then processes the transcript to extract memory and generate a wellbeing report. I need an experienced [login to view URL] developer to build the full automation pipeline connecting all components. The AI agent and system prompt are already written. You are building the plumbing, not the intelligence. --- **TECH STACK** - ElevenLabs Conversational AI - Twilio (outbound UK phone calls) - Airtable (user profiles, call records, memory storage) - [login to view URL] (all automation and orchestration) - Claude API / Anthropic (memory extraction + report generation) --- **DELIVERABLES — 5 [login to view URL] scenarios required** **Scenario 1 — Scheduled outbound call trigger** - Reads scheduled calls from Airtable - Retrieves caller profile + memory variables (11 fields) - Injects variables into ElevenLabs agent system prompt - Triggers outbound call via Twilio + ElevenLabs - Updates call status in Airtable **Scenario 2 — Post-call transcript processing** - Receives webhook from ElevenLabs when call ends - Sends transcript to Claude API - Extracts memory fields: forward thread, topics covered, overall affect, flags - Writes extracted memory back to Airtable caller record **Scenario 3 — Wellbeing report generation** - Triggered after each call - Sends full transcript to Claude Sonnet API with report prompt - Stores generated report in Airtable - Sends report by email via SendGrid (or similar) **Scenario 4 — Daily current affairs brief** - Runs each morning before first calls - Calls Claude API to generate a brief UK news + sport summary - Stores brief in Airtable for injection into same-day calls **Scenario 5 — Call duration hard stop** - Enforces maximum 7-minute call duration via ElevenLabs max_duration parameter - Logs any calls that hit the hard stop --- **AIRTABLE SCHEMA** You will design and build the following tables: - **Callers** — name, contact number, call schedule, interests, family members, sensitive topics, local resources, sports team - **Call records** — call number, date, duration, transcript, flag status, affect rating - **Memory** — forward thread, topics log, affect history per caller - **Reports** — generated report text, date, delivery status - **Current affairs** — daily brief, date --- **REQUIREMENTS** - All 5 scenarios fully working and tested - Minimum 3 end-to-end test calls completed with dummy user - Airtable schema documented with field descriptions - [login to view URL] scenarios documented with a simple flow diagram - Handover call (30 mins) to walk me through how to add new users and monitor calls - All API credentials set up and connected (I will provide keys) --- **WHAT I WILL PROVIDE** - ElevenLabs account access and agent already configured - Twilio account and UK phone number - Airtable account - All API keys (Anthropic, ElevenLabs, Twilio, SendGrid) - Full system prompt with all 11 variable names documented - Claude API prompt for memory extraction - Claude API prompt for report generation --- **IDEAL CANDIDATE** - Demonstrated [login to view URL] experience with HTTP/webhook modules - Previous work with voice AI platforms (ElevenLabs, VAPI, Retell or similar) - Experience with Airtable as a backend - Experience calling external APIs (Anthropic/Claude, Twilio) from [login to view URL] - Please include examples of similar automation projects in your proposal --- **TIMELINE: 7 days** This is a clean, well-specified project with all major components already decided. I am not asking you to design anything — only to build and connect what is specified above. Proposals that do not reference [login to view URL] and ElevenLabs specifically will not be considered.

Project ID: 40451471