AI assistant for searching internal regulations at ŘSD.

• Although ŘSD had a document management system, employees encountered several issues:
  

• The search functionality in the existing repository was limited to full-text search, which was insufficient for retrieving relevant information, especially given the variety of document formats.
  
• Extensive and complex legal and technical terminology made interpreting content more difficult.
  
• It was necessary to work with historical versions of documents (regulations) while maintaining access to up-to-date data.

  As a result, employees faced delays in finding correct information, the risk of misinterpreting rules, and reduced work efficiency.

An AI solution was created for semantic search and interpretation of ŘSD’s internal documents. Users can ask questions in natural language and receive answers with references to source documents, including versions valid at a specific date. The assistant uses OCR to process scanned files and runs on a hybrid architecture combining on-premise infrastructure with Azure AI services. Security, access rights, and full system control remain entirely with ŘSD.

Investment construction staff.
Technical support.
Public procurement and legislation experts.

3

Azure OpenAI (GPT-4o).
Azure App Service.
Azure Functions.
Semantic Search with vectors (pgvector / FAISS / Azure Cognitive Search).
OCR (Read API).
DOC BRO (custom framework for AI document processing).
Hybrid integration with internal servers.
RBAC access via Azure Entra ID.

Head of functional/operational unit

Full implementation

ŘSD internal team, trained and handed over know-how. Cloudfield available for support and further development on demand.

Internal knowledge base administrator. ICT system administrator. Documentation domain guarantor.

Internal file server or DMS (for documents). Azure subscription for running AI models. Access via Entra ID, VPN bridge, or hybrid connectivity. Budget for regular document updates.

• Faster document handling – semantically precise answers save employees significant time.
  
• Reduced errors – the AI assistant provides verified information with references to original directives.
  
• Easy process integration – users continued with their existing workflows, enhanced by AI capabilities.
  
• Security and control – ŘSD retained full management of the system without vendor dependency.

ŘSD successfully deployed an AI assistant for internal regulation search in collaboration with Cloudfield. The solution, built on the DOC BRO platform, enables fast and semantically precise document search, provides relevant answers in natural language, and ensures easy oversight of rule interpretation.

Tailoring the solution to internal processes. Domain knowledge (construction, public procurement). Full integration of permissions and security rules. Collaboration with ŘSD team and their openness to innovation.

Variety of documents and formats (PDFs, scans). Need to maintain version control and legal validity. Avoiding vendor lock-in and enabling independent operation.

Do not underestimate document analysis and structure – input quality determines AI output quality. Start with a small functional block and ensure answer interpretability (citations, references, versions).