

Fidoo is a Czech fintech company within the Direct group, specializing in corporate payment cards and digital expense management. For its new platform, we developed the backend for a card system and European Expense Management. At the same time, as the first team in the company, we introduced an AI-first development approach, which is now used across the entire Direct Fidoo organization.
TypeScript, Node.js, Azure Functions, CosmosDB, Azure Service Bus, Azure Event Grid, Claude Code, AI agents, AI SDLC
Fidoo was building a new platform that brings banking services, multibanking, expense management, reporting, and cash flow forecasting together in one place. Our task was to deliver its card component and European expense management, with an emphasis on the security and auditability required by fintech. The second goal was to move AI from a personal tool for individuals to a shared and repeatable way of working across the entire company.

Fidoo’s expense management solution has been years in the making and runs on its own in-house platform, whose documentation was outdated. We had to quickly get to grips with an extensive codebase. In addition, data from the card processor did not always arrive in the correct order, and several events related to a single transaction could arrive within a fraction of a second. Yet not a single event could be lost. The developers were initially skeptical about AI, and the first results proved them right. The generated code did not follow the platform’s conventions and failed code review.

We first store every incoming message from the card processor together with the raw data, and only then process it. Messages for a single card are queued sequentially in Azure Service Bus, and each processing run ends with a clearly named status, so audit and support teams can always see what happened to the message. For development, we built a team of specialized AI agents in Claude Code that understand the rules of the Fidoo platform. Before an agent submits its work, it must pass an automated check. Every error found by a human during review was turned into a new rule through automation. Thanks to this feedback loop, the agents were gradually refined to the point where they significantly accelerated the entire team’s work. We also helped design an AI SDLC process for product owners, analysts, developers, and testers. Today, our agents are available in the internal Fidoo Claude Kit marketplace, which is shared across the company.


Zdeněk Kalina
Software Architect, Direct Fidoo
“I have to admit that I approached the collaboration with a certain degree of caution. When working with an external agency, I was concerned that they would not show the same initiative and sense of responsibility toward the product as an in-house team—that they would simply do their job and leave it at that. And Fidoo is no trivial application. Our expense management module has been developed over many years, so getting to grips with the existing architecture and extensive codebase was far from easy. That makes me appreciate even more how the Qest team handled it. The initial phase was demanding; they needed to gain a thorough understanding of how the entire system worked and what it was built on. But once the team had acquired the necessary overview, they took full ownership of the project and began delivering at a very brisk pace. My initial concern proved completely unfounded. Qest approached the product with the kind of initiative and care I would expect more from our own people than from an external supplier. To me, that is the hallmark of a high-quality partner: the ability to tackle a complex challenge, learn quickly, and take ownership. They did an excellent job, and I can recommend them without hesitation.”
