Entrepreneur on Habr explains how he built a CRM for his auto repair shop with ChatGPT and Cursor

An entrepreneur from an offline business has described how, over two years, he turned AI from a curious toy into a working tool for his auto repair shop and related projects. With no technical background, he first built a simple CRM through prompts, then developed it into a web service, a Telegram bot and an Android app.

The first CRM in one night

The story begins in 2022, when the owner of an auto repair shop tried to find out whether AI could solve his very down-to-earth problems: messy records, weak bookkeeping, no analytics, and off-the-shelf CRMs that were awkward for mechanics who work as managers at the same time. After one conversation, the model suggested a simple move: instead of adapting to a ready-made product, build your own. That idea became the trigger. In a single night the author installed PyCharm, picked Python and SQLite on instinct, and ended up with a first working prototype in a single file of roughly two thousand lines.

"Make a CRM yourself with a clear interface."

What mattered most to him was not code quality but the fact that the system worked in a real business right away. In the morning he installed the program at the reception desk, and the staff figured it out without separate training.

Then came the path typical of early AI-assisted development: the monolith grew to six thousand lines, the context no longer fit into the model, and every new change got harder. Instead of asking "write me a CRM," the entrepreneur switched to "teach me to build a CRM in Python and Flask," then brought in Cursor and Claude, and gradually got to grips with architecture, VPS, DNS, certificates and PostgreSQL.

What GIPIX grew into

The raw desktop version eventually grew into the GIPIX CRM web service, used not only by the author's own team but also by acquaintances running offline businesses. He stresses that the interface was deliberately kept simple: the product is aimed not at IT people or office administrators but at mechanics aged 40+, who spend most of the day fixing cars and don't want to waste time on complex screens. In his words, this kind of practical adaptation proved more important than "beautiful" architecture or trendy design.

• 369 registered users
• about 95 active users per day
• 17 thousand client cards in the database
• online booking, cash register, warehouse and Telegram integration
• Android app, VIN data and AI analytics

Other products began to appear around the CRM in parallel. The author built GIPIX BIS AI, a system for modeling business ideas on a Canvas with about 150 modules, and then launched GIPIX VIN: a service with a Telegram bot and an Android app for working with VINs and license plate numbers. By his estimate, the latter already brings in about 80 thousand rubles a month with almost no promotion. Separately, he also built an offline CRM for a café for his own needs. The main CRM, however, is still in alpha testing, without classic monetization, and donations appeared only at users' request.

How he works with AI

The key idea of the article is that for the author, AI is not a replacement for the developer's profession but a way to build applied tools faster for a specific pain point. He describes a fairly disciplined process: he splits any task into stages, keeps a main prompt with the project context, gives the model the path to the relevant file, and asks it to first study the design, then critique the solution, and only after that make changes. This approach looks less like the magic of "do everything for me" and more like hands-on management of a junior but very fast digital worker.

This case shows clearly how the entry barrier to niche software development is changing. The entrepreneur still doesn't call himself a programmer, but he can already read code, understand basic architecture, deploy services and iteratively develop a product. What works best here is not the model's general-purpose power but deep domain knowledge: he knows how an auto shop runs, where money gets lost, why mechanics dislike complex CRMs, and which small details really save them time during the day.

What this means

Stories like this show that AI is increasingly becoming a layer between industry expertise and working software. For small and medium offline businesses, it is a signal: custom tools no longer require a full development team at the start, which means niche products can now be built by the people who have the problem themselves, even if the first version is rough but already useful.