CardBot

Telegram bot for generating marketplace product cards and AI product images for Wildberries and Ozon sellers.

CardBot turns a short product description and ordinary product photos into a ready-to-use marketplace card: SEO title, description, keywords, characteristics, and product visuals.

What It Does

The bot helps marketplace sellers create product listings faster by automating the repetitive part of card preparation:

• SEO-optimized product title
• product description
• keyword set
• structured characteristics
• AI-generated product images
• user balance and trial generation logic
• payment flow for paid generations

Why It Matters

Marketplace sellers often need many product cards, but preparing them manually takes time and requires copywriting, SEO, and visual production.

CardBot is a practical AI product that combines LLM text generation, image generation, Telegram UX, payments, and cost-aware generation logic into one workflow.

The project demonstrates not just an AI prompt, but a full product pipeline: user input, validation, LLM generation, image workflow, payments, balance accounting, and deployment.

Core Workflow

1. User sends a short product description and product photos.

2. The bot prepares the input for the AI pipeline.

3. LLM generates marketplace content:

   • title
   • description
   • keywords
   • characteristics

4. Image generation model creates product visuals.

5. The bot returns the generated card to the user.

6. Trial or paid generation balance is updated.

AI Pipeline

CardBot uses a multi-step AI workflow:

• DeepSeek / OpenRouter for structured text generation
• GPT Image for product image generation
• structured output for predictable card fields
• prompt engineering for marketplace-specific SEO format
• vision/image input workflow based on user-uploaded product photos

One of the key optimizations is combining several product photos into a single collage-like model input. This gives the image model more product context while reducing generation cost.

Product Features

• Telegram bot interface
• product card generation from text and photos
• SEO title, description, keywords, and characteristics
• AI product image generation
• trial generations
• paid generation balance
• payment integration
• generation history
• admin/product configuration through environment variables
• production deployment on VPS

Architecture

flowchart LR
    A[Telegram User] --> B[Telegram Bot]
    B --> C[Input Processing]
    C --> D[LLM Text Generation]
    C --> E[Image Generation]
    D --> F[Generated Product Card]
    E --> F
    F --> G[Telegram Response]
    B --> H[PostgreSQL]
    B --> I[Payment Provider]

Loading

Stack

• Language: Python
• Bot: Telegram Bot API
• Database: PostgreSQL
• AI / LLM: OpenRouter, DeepSeek, GPT Image
• Payments: payment provider integration
• Infrastructure: Linux VPS, systemd, GitHub Actions
• Architecture: async bot workflow, structured generation pipeline, balance/payment logic

What This Project Demonstrates

• building an AI-powered product end-to-end
• LLM integration in a real user workflow
• AI image generation pipeline
• marketplace/e-commerce automation
• cost-aware AI product design
• Telegram bot UX
• payment and balance logic
• production deployment and operation

Repository Notes

This repository is part of my public portfolio and demonstrates the product architecture and implementation approach behind CardBot.

Sensitive production credentials, API keys, payment secrets, and environment-specific configuration are not stored in the repository.

Status

MVP launched as a Telegram bot.

Public bot: @CaardMakerBot

Related Links

• Product / brand: alterega.ru
• Author: Egor Dushenko