She built an AI receptionist for her brother's shop
A developer built an AI voice receptionist using Claude and Vapi for a mechanic shop losing thousands monthly from 100+ missed calls per week.
A Chicago developer named Kedasha Kerr just built an AI-powered phone receptionist called Axle — and the story behind it is the kind of practical AI automation most people dream about but never try.
Her brother runs Dane's Motorsport, a luxury mechanic shop. The problem? He was losing thousands of dollars every month because he couldn't answer the phone while working under vehicles. Over 100 calls per week went unanswered — each one potentially a $450 brake job or a $2,000 engine repair walking straight to a competitor.
So she built an AI that answers every call, 24/7.
How Axle actually works
Axle isn't a generic chatbot reading a script. It's a RAG-powered voice agent (RAG means the AI pulls answers from a real knowledge base instead of guessing) that knows everything about the shop — 21 documents covering services, pricing, hours, warranty policies, even loaner vehicle availability.
When a customer calls, here's what happens behind the scenes:
1. The phone rings → Vapi (a voice AI platform handling 300M+ calls) picks up and converts speech to text using Deepgram
2. The question gets matched → The customer's question is converted into a number pattern (an "embedding") using Voyage AI and matched against the shop's knowledge base in MongoDB
3. Claude writes the answer → Anthropic's Claude Sonnet generates a natural response — but only from the knowledge base. If it doesn't know, it says so
4. The customer hears a voice → ElevenLabs converts the text to a calm, natural voice named "Christopher"
The whole loop takes under a second. A customer asking "How much is an oil change?" hears: "$45 for conventional, $75 for synthetic. Includes oil filter, fluid top-up, and tire pressure check. Takes about 30 minutes."
The 20 voice tests that made it sound human
Building the AI brain was one thing. Making it sound right on the phone was another challenge entirely. Kerr ran 20 different voice tests before landing on the right combination:
• Responses capped at 2–4 sentences max — nobody wants a lecture on the phone
• Prices spoken naturally — "forty-five dollars" not "$45.00"
• All filler phrases removed — no "Certainly!" or "Great question!"
• Short sentences only — no markdown formatting leaking into speech
When the AI can't answer a question, it doesn't hallucinate. It collects the caller's name and phone number for a callback — logged automatically in a database her brother can check between jobs.
The tech stack anyone can replicate
What makes this project especially interesting for our readers is that it uses all off-the-shelf tools. No custom machine learning. No training. Just smart wiring:
• Vapi — phone infrastructure (speech-to-text + text-to-speech)
• Claude Sonnet 4.6 — the brain that generates answers
• MongoDB Atlas — stores the knowledge base with vector search
• Voyage AI — converts text into searchable number patterns
• ElevenLabs — natural voice output
• FastAPI + Python — glues everything together
• Ngrok — for local development and testing
Kerr's core design principle: "Don't use a raw LLM for a business-specific voice agent. Ground it in a real knowledge base, constrain it to only answer from that base."
Who should pay attention
If you run a small business — a salon, a dental office, a repair shop, a law firm — and you're losing clients because nobody answers the phone, this is your blueprint. The tools Kerr used all have free tiers or affordable pricing.
If you're a developer looking for a weekend project with real impact, Kerr's full tutorial walks through every line of code. Part 2 will add calendar integration for direct appointment booking, and Part 3 adds SMS notifications.
If you're exploring AI automation, this is the clearest example yet of how off-the-shelf AI tools can solve a real business problem — not a hypothetical one. A mechanic shop was bleeding revenue. Now an AI picks up every call.
What happens next
This is Part 1 of a three-part series. The upcoming features — direct appointment booking via calendar integration, SMS callback alerts, and a management dashboard — will turn Axle from a smart answering machine into a full AI receptionist that books, confirms, and follows up.
The post hit Hacker News with strong engagement, with commenters noting that this exact pattern — voice AI + RAG + domain knowledge base — could be replicated for virtually any service business.
Related Content — Get Started with Easy Claude Code | Free Learning Guides | More AI News
Stay updated on AI news
Simple explanations of the latest AI developments