Demo: https://www.loom.com/share/e4cdc5b60a3949c2bf5c32bf5caca297

RoadBuddy AI

A Telegram chatbot that teaches driving theory with real street photos from your own city, one quick quiz at a time.

Features

• Hyper-local quizzes: Utilizes Mapillary street-level images from the user’s city.
• Multimodal LLM: Mistral AI generates a multiple-choice question with 3 options and a one-sentence explanation.
• Infinite learning: You can process an infinite number of completely unique questions.
• Zero setup: Everything runs directly in Telegram—no additional apps required.

🙌 Made by:

Tech Stack

• Bot: Telegram Bot API with Telegraf SDK
• Images: Mapillary REST API
• LLM: Mistral AI
• Backend: Express on TypeScript
• Storage: PostgreSQL DB

Architecture explanation

Dependencies

• Docker
• Ngrok

Getting Started

1. Clone the repo and run these commands

   git clone https://github.com/yvann-ba/RoadBuddy-AI
   cd RoadBuddy-AI
   cd bot/
   npm i
   cd ..

2. Setup ngrok

In another terminal please run:

ngrok htpp 3000

Copy the URL highlighted in the image above to set up your environment variables later.

4. To create your own bot, simply follow these steps

   For any questions or assistance, feel free to reach out to me:

   • LinkedIn: Andrii Syvash
   • Email: asyvash.work.it@gmail.com

5. Create .env file using this template

   POSTGRES_DB=telegram_bot
   POSTGRES_USER=postgres
   POSTGRES_PASSWORD=postgres
   DATABASE_URL=postgresql://${POSTGRES_USER}:${POSTGRES_PASSWORD}@db:5432/${POSTGRES_DB}
   
   TELEGRAM_TOKEN=
   BOT_PORT=3000
   WEBHOOK_URL=

   Paste the ngrok URL you copied earlier into the WEBHOOK_URL variable. Also, paste your Telegram token into the TELEGRAM_TOKEN variable.

6. Almost ready to start!

   If you followed all the steps correctly, run the final command to start the project:

      ./setup.sh start

Potential bussiness model:

• We can limit daily messages to 10. After that, users will need to purchase a subscription.
• A perfect solution for passing the driving exam, as you will practice with real images from your city.

Team:

Andrii Syvash
Ismael Mehdid
Jan Matter
Yvann Barbot