A Mobile Phone Out of Almost Nothing: How AI Helped Build a Phone from Parts

⚡ The Gist in 5 Seconds - The author built a working GSM phone out of modules bought on marketplaces, using free versions of DeepSeek, Gemini, and ChatGPT. - The device can make and receive calls, and display signal strength and the time on screen. - It's an experiment, not a smartphone replacement: the functionality is primitive, with no call log or other conveniences. ### 🔍 What Was Found In an article on Habr, journalist Viktand described in detail how he built a phone "out of almost nothing." The foundation is a SIM800C radio module (HW-537 board), an STM32L051C8T6 microcontroller — a choice suggested by Gemini — and an SSD1306 OLED display with a 128×32 resolution. All components were bought on marketplaces. The module is controlled via UART with AT commands, and the code was generated by neural networks: DeepSeek provided the initial firmware, while ChatGPT later handled refactoring and added on-screen "eye candy." The circuit was assembled on a breadboard. At first, power was supplied via 5V USB, and the radio module kept rebooting. The cause was peak currents of up to 2A that the SIM800C draws when communicating with a base station. The capacitors couldn't cope, so a 75 mAh lithium battery salvaged from a Fix Price toy had to be added. Later the power circuit was reworked: charging was built in, and for a 320 mAh battery Gemini suggested an 18 kΩ resistor — the actual charging current came out to about 180 mA. The keypad was assembled from tactile buttons. The author wrote the button-reading code himself, and DeepSeek integrated it into the main program. The result is a fully working device: it receives incoming calls (one of 19 built-in melodies plays through the speaker), allows dialing numbers, and shows signal strength and the time. The case was drawn in OpenSCAD, also with Gemini's help. ### 💡 Why It Matters The article shows how AI speeds up prototyping. The author deliberately didn't dig into the radio module's protocols or spend time studying documentation — he simply asked the neural networks questions and immediately got wiring tables, code, and advice. For beginners, this is a ready-made entry scenario into the world of microcontrollers: even without deep knowledge, you can build a working GSM phone using free tools. That said, the project remains an honest experiment: a regular feature phone is cheaper and easier to just buy at a mobile store. ### 🧩 Context The author used only free DeepSeek, Gemini, and ChatGPT in limited mode, with no particular system for choosing a neural network — he simply switched whenever one started "acting up." The Habr article was published on September 27, is a 7-minute read, and has racked up 31,000 views. The project went through several stages: from testing the basic circuit to built-in charging and a full keypad, with final touches adding two more 2000 µF capacitors. It's a vivid example of the DIY and AI symbiosis that anyone can replicate with their own hands.

🔗 Read on habr.com