I Built an AI Receptionist for a Luxury Mechanic Shop – Part 1
3 hours ago
- #RAG Pipeline
- #AI Receptionist
- #Mechanic Shop
- The author built an AI receptionist named Axle for their brother's luxury mechanic shop to handle missed calls and prevent lost business.
- The AI receptionist uses Retrieval-Augmented Generation (RAG) to provide accurate answers based on a scraped knowledge base from the shop's website.
- The knowledge base includes service types, pricing, policies, and other details, stored in MongoDB Atlas with vector embeddings for semantic search.
- Claude (Anthropic) generates responses strictly from the knowledge base to avoid hallucinations and maintain accuracy.
- Vapi was chosen as the voice platform to handle telephony, speech-to-text, and text-to-speech, with a FastAPI webhook server for processing queries.
- The system logs all calls and callbacks in MongoDB for tracking and analysis, turning phone interactions into valuable data.
- Voice tuning was critical, with careful selection of a natural-sounding voice (Christopher from ElevenLabs) and rewriting prompts for conversational delivery.
- The stack includes Vapi, Ngrok, FastAPI, MongoDB Atlas, Voyage AI, Claude, and Python for integration.
- Future enhancements include calendar integration for booking appointments, SMS notifications, and a dashboard for managing callbacks.
- The key insight is to ground the AI in a real knowledge base and design a robust fallback flow for unanswered questions.