A luxury mechanic shop owner loses significant revenue monthly due to missed calls. His brother built Axle, a custom AI receptionist, to answer calls, access pricing, and collect callbacks. Axle uses Retrieval-Augmented Generation (RAG) to prevent hallucinations, pulling accurate data from the shop's website into MongoDB Atlas via Voyage AI embeddings.

When a customer calls, Vapi handles the call, sending queries to a FastAPI webhook. The webhook routes the question to the RAG pipeline, which uses Claude for grounded responses. If the AI can't answer, it collects a callback.

The system logs all interactions, turning the phone system into valuable data. Voice tuning involved selecting a natural-sounding ElevenLabs voice and rewriting prompts for conversational delivery. The stack combines Vapi, MongoDB, Voyage AI, and Claude.

Axle currently answers questions and takes callbacks.