RAG AI Voice Customer Support Agent (Supabase + Gemini + ElevenLabs)

This workflow creates a fully automated RAG-powered AI voice customer support agent. It integrates ElevenLabs for conversational voice input/output, Supabase for vector database storage and semantic search, Together AI for embeddings, and Google Vertex Gemini for intelligent response generation. When a user speaks, ElevenLabs sends the transcribed question to a webhook, which triggers embedding, context retrieval, LLM processing, and response back for voice synthesis—enabling natural, context-aware conversations without human intervention.

The process breaks down into key steps: (1) Webhook receives user question from ElevenLabs; (2) Embed message using BAAI/bge-large-en-v1.5 via Together API; (3) Search top 5 relevant chunks from Supabase knowledge base via RPC; (4) Aggregate retrieved contexts; (5) Use LangChain's Basic LLM Chain with Gemini 2.5 Flash to generate human-like answers strictly from context; (6) Respond instantly to webhook for ElevenLabs to vocalize.

Benefits include 24/7 scalable support, reduced costs over live agents, hyper-personalized responses via RAG, and seamless voice interface. Ideal use cases: e-commerce helpdesks, tech support hotlines, virtual receptionists, or any service needing instant, knowledgeable voice interactions. Setup requires API keys for integrations and a pre-populated Supabase vector store.