Raiffeisen Tech Poland From ideas to AI solutions – scaling innovation across RBI Group teams through a hackathon

What truly defined the hackathon was not the element of competition, but its hands-on, workshop-driven nature. Participants went through the full lifecycle of building an AI product — from ideation, through design, all the way to a working demo — while strengthening collaboration, knowledge sharing, and a practical approach to fast-paced delivery.

Eighteen participants from Raiffeisen Tech Poland and the Vienna-based Group Risk Data & Regulatory Reporting team, split into six three-person teams, worked on real-world IT solutions. Along the way, they developed their AI capabilities and applied a modern, product-driven approach to building technology. The event encouraged collaboration and sharing knowledge and experience.

The hackathon was designed around two clear objectives — building AI capability across teams and delivering real, usable products.

The hackathon kicked off with check-ins and team formation, but the real excitement began with the random assignment of both the business problem and the technological approach. Teams worked either with GitHub Spec-Kit or BMAD, taking on an end-to-end challenge — from concept, through technical design, all the way to a working demo.

Each solution had to meet functional requirements, maintain high code quality, and align with RBI’s technology standards, ensuring that ideas were not only innovative, but also production-minded from the very beginning.

The first use case focused on Data Quality and aimed to simplify the work of Data Stewards through a simple and more intuitive interface for managing Data Quality Indicators and their lifecycle, as well as the automation of key processes.

The second use case covered AI-powered license management — a platform designed to support contract analysis and license record management. The solution was intended to automatically extract key information from contracts via AI and enable users to ask questions against the beforehand extracted contracts, in natural language , supporting compliance and self-service capabilities.

Teams operated within RBI’s real technology landscape — no sandbox shortcuts, real enterprise constraints.

Projects were assessed from multiple perspectives, capturing both the quality of the product itself and the way it was built and presented. The focus was on user experience, technical excellence, and the overall clarity of delivery. Judges looked at the functionality and usability of each solution, the quality of its architecture and code, and the effectiveness of the applied AI mechanisms. Security and the interpretability of results were also key factors, ensuring that solutions were not only innovative, but also reliable and transparent.

Equally important was the ability to clearly communicate business value within a short, 5-minute demo — a challenge itself in a fast-paced hackathon environment.

All of this took place under the watchful eye of the jury:

• Thomas Killar – Group IT Delivery Non-Client Facing, Raiffeisen Bank International
• Philipp Pfneisl – Group IT Delivery Non-Client Facing, Raiffeisen Bank International
• Marcin Przewoźnik – Raiffeisen Tech Poland
• Danijel Laklija – Group Risk Data & Regulatory Reporting, Raiffeisen Bank International 
• Jurgen Stadler – AI Coach, Raiffeisen Bank International
• Robert Ruzitschka – AI Coach, Raiffeisen Bank International

During the hackathon, teams showcased a broad spectrum of approaches to building systems around DQI challenges and document intelligence—ranging from lightweight MVPs and optimized workflows to advanced enterprise-grade architectures and AI-driven platforms.

Team D presented a solution tightly integrated with GitHub, leveraging pull request workflows to manage changes in DQI. They extended the core system with a dashboard layer, search capabilities, and a basic approval pipeline, creating a pragmatic, developer-centric change management loop.

Team E focused heavily on usability and data operations. Their system emphasized operational efficiency, delivering a robust set of features for filtering, editing, previewing, as well as Excel import and YAML generation—essentially enabling fast, structured data manipulation through a highly accessible UI layer.

Team F1 pushed furthest into governance and enterprise process design. They combined business-role modeling with automated PR generation, AI-driven risk analysis, and natural language–controlled bulk updates. The result was a highly comprehensive, end-to-end enterprise workflow model with strong emphasis on control, auditability, and automation.

In the contract intelligence track, Team B delivered a document extraction system for licensing contracts, enhanced with confidence scoring and a “talk to your data” assistant, enabling interactive exploration of extracted insights.

Team C evolved this direction into a production-grade platform, introducing OCR pipelines, PostgreSQL-based persistence, containerized deployment, versioning mechanisms, and a context-aware AI chat interface—bringing the concept closer to a fully operational enterprise system.

Finally, Team A contributed a critical process insight around the BMAD methodology. Their observation was that while this highly detailed, specification-heavy approach delivers strong value in longer, structured delivery cycles, it becomes overly heavy and inefficient in the compressed timeframe of a hackathon environment.

The score differences between the top teams were minimal (under 5%), clearly demonstrating an exceptionally balanced and high-level performance across all participants. Ultimately every team was a winner, with valuable lessons learned that will shape future projects.

Two days of intense work, a multitude of ideas, and even more inspiring conversations have demonstrated the immense potential within our teams. Innovation emerges where technology, the courage to experiment, and a well-defined business problem intersect.

The combination of AI, risk data considerations, strong engineering practices, and cross-team collaboration quickly translates into real, high-value solutions. The hackathon was also an opportunity to develop future-ready skills—from data literacy and secure design to building AI-ready products.

Teams that delivered working MVPs the fastest typically relied on a SpecKit-driven approach, which enabled rapid translation from specification to implementation and iterative product improvement through agent-based workflows. Agent-assisted testing and code review also proved highly effective, as did parallel work across branches—especially in cases where the specification was clearly defined and stable. The strongest results emerged when the problem was well described and did not require continuous clarification during implementation.

At the same time, several important limitations became apparent. The BMAD approach turned out to be too heavyweight for a two-day hackathon, leading to iterative loops in PRD refinement and validation, while also making it difficult for multiple contributors to collaborate without a shared single source of truth. In cases of weak documentation, AI-generated code exhibited hallucinations and inconsistencies, and debugging still required significant manual intervention.

The key takeaway from the entire hackathon was that the quality of the initial documentation and the precision of the specification had a direct and dominant impact on the quality and consistency of AI-generated code—far greater than the choice of tools or implementation frameworks themselves.

What we learned from two days of building with AI.

A heartfelt thank you to the organizers, mentors, coaches, jury, and participants for creating a space to learn, experiment, and collaborate across international teams. It was this sense of collaboration, the exchange of perspectives, and the shared approach to problem-solving that truly became the greatest value of this hackathon.