The developer created a Telegram bot with a Mini App that manages grocery shopping and meal planning. The system uses AI agents to generate menus and recipes based on available ingredients, budget, and caloric goals, while Python code enforces strict constraints on calculations and inventory.
The project was completed in 4 days, resulting in approximately 4,400 lines of Python code and 1,300 lines of HTML/JS. The developer implemented 498 tests to ensure system stability. The bot integrates with Yandex Lavka, allowing users to assemble shopping carts automatically, though payment remains a manual step to prevent errors.
The architecture relies on a 'model proposes, code verifies' principle. AI agents handle creative tasks like menu generation, while the code validates caloric intake, budget limits, and packaging requirements. A shared STATE.md file serves as a memory log for agents, documenting technical constraints and 'traps' identified during development.
The developer noted that relying solely on mock tests for API interactions is insufficient, as real-world factors like specific darkstore inventory and address-based pricing discrepancies caused initial failures. The system requires manual verification of the cart before checkout to mitigate risks associated with unofficial API usage.
