World Bank: AI advisory for Clean Air Programme.

The problem

One of Europe's largest thermal-modernisation programmes, run across a complex, multi-vendor, multi-institution IT landscape (the national NFOŚiGW fund, 16 regional funds, state registries) — with widely varying IT maturity and a high share of errors in applications.

What I built

Architecture advisory documented in two reports:

• An IT architecture report — a diagnosis of the programme's bottlenecks from an IT perspective (including the high share of faulty applications in the GWD system), with recommendations on data validation, contextual help, integrations and modernisation.
• An AI and automation memo — use-case analysis (intelligent application validation, OCR, automated document verification), model selection (cloud: Vertex/Claude, Azure OpenAI, and on-premise options including the Polish models Bielik and PLLuM), GDPR and AI Act compliance, a phased implementation roadmap and a target architecture.

The outcome

Two comprehensive advisory reports with prioritised recommendations — from quick wins in data validation and a central OCR service, to a central AI system for preliminary application verification and on-premise infrastructure integrated with the new GWD/NSOW — together with a phased delivery model and target architecture.