Alfa-Bank turns a tester's idea into an AI stub generator for QA teams

An Alfa-Bank tester has built an internal AI tool that turns a text description into ready-made API stubs for service testing. The project started as an evening experiment inside the AI community and grew into an MVP, a dedicated domain and in-bank testing.

How the project started

The story began in spring 2025, when testing engineer Stas Zaitsev decided to dig deeper into neural networks and see how they could be used in everyday QA work. After several experiments, including a Telegram bot for assessing English, he joined the internal AI community built around the AlfaGen platform.

At one meeting, participants were shown a list of useful ideas that teams had not yet found time to tackle. One was a task familiar to any tester: automatically creating API stubs.

The idea was practical and down to earth. When the frontend is ready but the backend is still in development, testing often stalls because the service returns no working responses. Stubs solve that problem, but preparing them takes time, manual configuration and an understanding of the infrastructure.

The business requirements for the new agent described a simple scenario: the user sets the conditions, the system asks clarifying questions and produces stub code that can go straight into tests without lengthy preparation.

How the generator works

Zaitsev built the first version very quickly in AlfaGen's "Prompt Workshop": he configured a scenario that generated stub code from a description. But that format proved inconvenient for real work and scaled poorly to colleagues without a technical background. A long chat dialogue, a manual launch and the need to understand the details made it more a demonstration of the idea than a tool for the daily work of QA teams. That is why a second version, with a separate interface and automatic launch, was needed.

"The project was supposed to take 3 months, but I didn't know that and

did everything in one evening."

In the second version, the author built a full agent in three parts: WireMock to run the stubs, an API layer on Kotlin and Spring AI to connect to the AlfaGen web interface, and a UI usable without deep programming knowledge.

The user simply describes the desired scenario in a chat window, for example which responses to return for different userId values, whether a delay is needed and when to return a 404 error. The system then goes through several steps:

• determines the stub type, REST or SOAP
• selects the right HTTP method
• calculates how many separate stubs are needed for the different conditions
• generates the code and checks it for errors
• deploys the result in WireMock

According to the author, the whole cycle takes 10–15 seconds, after which the stub can be plugged into a working environment. For those who need full control, the interface also keeps a manual mode where parameters are set individually.

That makes the tool useful both for experienced engineers and for testers who do not want to spend time building and launching mocks by hand. In practice, the agent removes almost all the routine mechanics from the user.

From MVP to launch

The evening prototype quickly grew into a two-week MVP and then into a full internal product. After a presentation at an AlfaGen platform demo, other teams at the bank became interested in the tool, and the developer joined a group that works on QA agents full time.

The project was then brought to production quality: the team added a chat mode, per-project folders, stub groups with logic, fixes based on colleagues' feedback and a more convenient user flow. By the end of November, the team had closed the critical issues and begun preparing for a broader launch.

The solution received a dedicated domain, access controls were configured, the frontend was rewritten to use the bank's internal libraries, and documentation was prepared. The generator is now being tested by focus groups inside the bank, while a local version is already used on real tasks in products for small business.

In essence, it is an example of how an internal AI platform turns an individual engineer's pain point into an applied service for an entire organization.

What this means

Alfa-Bank's story shows that useful AI tools for development can emerge not only from dedicated R&D teams but from the bottom up, out of testers' concrete pain points. If such projects are quickly taken through an interface, documentation, internal standards and a real pilot, they stop being a demonstration of what LLMs can do and become a working tool that saves the whole team hours and speeds up the arrival of new services in testing.

For enterprise AI, that matters more than any polished demo stage.