AI for CRM: development without IT experience

How AI Helps a Non-IT Entrepreneur Build a CRM for Offline Business

An entrepreneur with no IT experience used AI to develop a CRM for an auto repair shop. Instead of implementing off-the-shelf systems, he created a custom solution in Python with SQLite. In 3 hours, he had an executable file, GIPIX_v0.1.exe, with modules for records, cash register, inventory, and admin panel. The mechanics mastered the interface without training—the key to success was simplicity for users over 40.

The tech stack was chosen minimally: Python for logic, SQLite for a local database. AI generated code based on prompts like 'make a CRM for mechanics.' The result: 2,000 lines in a single main.py, but functionality covered client records, cash register, and inventory.

Evolution from Monolith to Web Version

The monolith grew to 6,000 lines, and the context no longer fit into ChatGPT prompts. Transition to learning: AI explained the Python + Flask architecture. Over 3–4 months, the author grasped the principles and learned to read code.

In 2024, Cursor AI accelerated web version development. Issues with VPS, DNS, and PostgreSQL were resolved via Claude. The outcome—GIPIX CRM in production: Telegram bot, online booking with generated websites, client accounts.

Not everything worked immediately: the offline version caught on better than the early web version. Adaptation for mechanics—interface simplicity, auto-text correction, hour norms.

Functionality and System Metrics

The system evolved iteratively through prompts. Current capabilities:

• Client records with smart funnels (follow-up after 90 days, bonuses)
• Cash register with document flow
• Inventory with photo integration and cross-references
• Online booking: unique website + account for each client
• Telegram bot and MAX (notifications)
• VIN data and AI analytics
• Offline version on Electron
• Android app with call capture

Metrics as of March 2026:

• 369 registered users
• ~95 active daily
• 17k clients in the database
• Donations ~40k rubles (no advertising)
• No monetization, alpha test

Prompting Methodology for AI Development

Development proceeded in iterations: tasks were broken into stages, using a base prompt with project context. Example for UI edits:

Need to remove the 'Payment' button and replace it with 'Accept Payment' with a pulsing effect @index.html.
1. Study the design and colors.
2. Work on positioning and usability.
3. Critique, suggest an alternative.
4. Analysis in 3 lines.
After 'start'—changes, #key comment in 5 words.
After 'finish'—git and FTP instructions.

The approach scaled to other projects: GIPIX BIS AI (Canvas with 150 modules based on Osterwalder), GIPIX VIN (auto database by VIN with Telegram bot and Android app, revenue ~80k rubles/month), offline CRM for a cafe.

Key Takeaways

• AI replaces a junior developer for non-IT users: generates code, explains architecture, debugs.
• Focus on users: interface for mechanics without IT skills.
• Iterative development through detailed prompts saves time.
• Scale from local exe to cloud with PostgreSQL and integrations.
• Metrics show viability: 95 DAU without marketing.

— Editorial Team