As a car mechanic and shop owner, I got tired of answering
the same diagnostic questions over and over. "Why is my
Check Engine light on?" "Why does my car shake at 60 mph?"
"Why won't it start in the cold?"
So I built an AI-powered diagnostic tool for my auto service
website. It's a conversational chatbot that asks follow-up
questions, analyzes symptoms, and gives users a preliminary
diagnosis — all before they even call the shop.
How it works
The bot (I named him "Paweł") uses an LLM to:
- Accept a free-text description of the problem
- Ask targeted follow-up questions
- Suggest likely causes ranked by probability
- Recommend next steps (DIY fix or visit a mechanic)
What surprised me
- Users describe problems much better when talking to a "mechanic" than filling out a form
- The AI catches patterns humans miss (e.g. correlating cold weather + hard starts + blue smoke = likely valve seal issue)
- It reduced "I don't know what's wrong" walk-ins by ~30%
Tech stack
- Frontend: embedded iframe, plain JS
- Backend: Railway-hosted Node.js service
- LLM: Claude API for natural conversation flow
You can try the live demo here:
👉 
Curious if anyone else has built domain-specific AI
assistants for non-tech businesses? Would love to compare
notes.
Top comments (0)