How Masha Leshchinskaya Saved Her Pet Project from Death: A Device Booking System for AI

Masha Leshchinskaya, Head of QA at Surf, shared the story of a pet project that survived unlike hundreds of others. The device booking system has been running for three months. And this is not thanks to revolutionary technology or venture funding, but thanks to one simple solution: minimizing maintenance costs during the design phase.

Why Pet Projects Die

Most side projects live for a week or two. A developer gets excited about an idea at night, quickly launches an MVP over the weekend, pushes it to GitHub, shares it on social media. But then reality hits: you have to monitor logs at 2 AM, fix production errors, add features based on early users' requests. All in your free time, without deadlines, without budget, without analysts. Enthusiasm drops linearly over time, and the project gets frozen in a GitHub archive.

Masha faced the same trap but did things differently: instead of perfect architecture with maximum flexibility, she immediately chose minimal maintenance as her primary criterion.

How to Design a System for Minimal Maintenance

The device booking system at Surf solves a concrete problem: QA engineers and testers need to book physical devices (smartphones, tablets, IoT) for testing. Usually, it's either a Google Sheet with conflicts or an expensive enterprise tool like TestRail. Masha chose a functional minimum and automated everything possible:

• Automatic device release on timeout—a device is booked for exactly the allocated slot, then automatically releases itself
• Notifications through existing channels (Slack, Telegram)—no separate app needed
• Database synchronization is fully automated, no manual updates
• Minimal custom integrations—uses ready-made APIs (Slack, Google Calendar)
• Logging and monitoring self-diagnose, with alerts only for real issues

Result: the project requires full attention for only 15 minutes once a week, not daily monitoring.

How the Pet Project Became a Testing Ground for AI

Masha used the system as a real testbed for AI and LLM experiments. For example, an NLP bot that understands natural language requests ('I need an iPhone 15 for an hour from 3 to 4 PM') and books the device without any UI clicks. Or an AI agent that suggests alternative devices if the needed one is busy.

A classic research bot in a Jupyter notebook gets forgotten. A bot in a production system lives longer: it learns from real errors, from real user behavior, not from synthetic datasets in documentation.

What This Means

A long-lived pet project is not magic or luck. It's the result of sober design: minimize maintenance pain, automate routine, choose a practical use case. And if the project becomes a real testing ground for AI experiments, it stops being a hobby—it transforms into a tool worth maintaining.