Czech Ministry of Labour: An Ai-Generated Data Dictionary Was Built in Just Days — Capturing 30 Years of Data
Dozens of databases, more than 13,000 tables and three decades of records. Manually mapping it all would have taken years — yet the Ministry needed clarity now to improve digital services and make real use of AI. The solution? Generative AI capable of making sense of a highly complex system and organizing it in just a few days.
columns across more than 13,000 tables
less data after removing duplicates
hours to process a single schema instead of weeks
Challenge
Quickly Make Sense of Dozens of Databases and Create a Data Dictionary Ready for AI
How much data can an institution gather in 30 years? At Czech Ministry of Labour and Social Affairs (MPSV), it means dozens of databases, more than 13,000 tables, and over 100,000 columns. Every system and team left its own trace — different names, structures, and descriptions.
For new employees, onboarding took weeks; a full understanding of the data landscape was something only a few experts possessed.
Meanwhile, business teams (policy departments, the Labour Office, the Social Security Administration) needed a practical way to understand what data the organization actually has.
The push to modernize public IT, rising transparency requirements, digital services, and demand for analytics all made one thing clear: institutions need to know their data, understand it, and be able to share and connect it securely.
But mapping such a large and complex environment the traditional way would take years of painstaking manual work — and constant upkeep just to keep it current.
Solution
Central Data Catalog — Built in Days Through Collaboration Across Data, Domain, and AI Experts
The Ministry of Labour and Social Affairs partnered with Adastra and, together with Microsoft, ran a pilot hackathon.
Data analysts, AI specialists, and internal domain experts worked side by side — combining technical skills with deep subject knowledge.
Using large language models (LLMs), the AI team was able to automatically:
- Classify hundreds of tables and thousands of columns as either “business” or “technical”
- Detect and deduplicate dozens of different names referring to the same data
- Create “meta-columns” that standardize and describe data within each domain (e.g., social policy, employment
The result was a business data dictionary — a central catalog that shows:
- what information the Ministry holds,
- where it is stored,
- and how it’s linked.
It is designed for business users but maintains a precise connection to the physical data model, making it an effective bridge between business teams and data analysts when refining requirements.
Work that would have taken years by hand can now be completed by LLMs in hours to a few days.
The output was validated through random sampling of hundreds of newly created terms/columns and their descriptions.
These were manually reviewed and graded on a school-style scale; the average score ranged between 1.5 and 1.7 (1 = best).
How the data dictionary helps in practice
Before:
Onboarding new employees into the Ministry’s data environment meant weeks — sometimes months — of piecing together context and relationships across systems and tables.
Now:
Newcomers can immediately see which databases hold specific information, how it connects, and what it’s used for — significantly reducing onboarding time.
Before:
Business teams often didn’t know what data existed.
The physical data model was too technical and complex to be useful.
Now:
Business users can explore their domain’s logical data items in Power BI — including full-text search — giving them a clear view of what data they can work with.
Impact
A Business Data Dictionary Ready for AI Tools, Analytics, Reporting, and Legal Data Records
A clear, unified data dictionary opens new possibilities for the Ministry of Labour and Social Affairs:
- A foundation for future AI tools — for example, chatbots that let users query data in natural language without needing to understand database schemas.
- Easier system modernization — identifying redundant columns and structures means only relevant, mapped items need to be migrated when systems evolve.
- Faster analytics and reporting — enabling use cases such as predicting benefit payments, modelling client return patterns, or recommending job openings.
- Support for regulatory requirements — the Ministry can clearly document which data it manages and where it is stored.
For the Ministry, data is a key asset for future service development, modernization efforts, and AI projects.
By mapping and organizing its data into a shared dictionary, the Ministry has taken an important step toward adopting modern technologies more efficiently and with less friction.
