CZR: Large Language Models Empower Public Contract Analysis, Semantic Search Unlocks New Value of Government Data

The CZR project is an open-source public contract data analysis system focusing on data from Slovakia's Central Contract Registry (CRZ). It builds a complete data processing pipeline using semantic search and large language model technologies to enable intelligent downloading, processing, and analysis. It provides an innovative example for government data transparency and intelligent analysis, helping to improve the efficiency of public fund usage and anti-corruption effectiveness.

The CZR project's data comes from Slovakia's Central Contract Registry (CRZ), a platform that requires registration of public contracts exceeding the threshold (including government procurement, engineering, service outsourcing, etc.), covering central to local governments. However, the original data has pain points: traditional keyword search cannot meet deep analysis needs, and manual screening is inefficient and prone to missing key information. The CZR project aims to solve this problem and realize automated processing and value mining.

The CZR system architecture consists of three layers:

1. Data Collection Layer: An intelligent web scraping module automatically traverses the CRZ directory, processes formats like PDF/Word/HTML, extracts metadata (parties to the contract, amount, etc.), and supports incremental updates;
2. Data Processing and Storage Layer: Parses documents into structured text, extracts key fields using rules + machine learning, and stores them in a vector database after unifying the format;
3. Intelligent Analysis Layer: Semantic search understands query intent (e.g., "IT infrastructure upgrade" matches relevant contracts), and large language models are used for summary generation, risk identification, classification labeling, and multilingual processing.

The social value of the CZR project is reflected in:

• Transparency Enhancement: Journalists/researchers track the contract history of projects or companies to discover conflicts of interest or anomalies;
• Efficiency Optimization: Government procurement departments refer to historical data to avoid overpricing;
• Academic Research: Scholars conduct empirical research on government procurement efficiency, competition level, etc., based on structured data;
• Citizen Supervision: Ordinary citizens understand the direction of government funds through search and participate in supervision.

The general insights from the CZR project include:

• Data Standardization: Structuring government data is a prerequisite for intelligent analysis, and a unified standard needs to be established in the collection phase;
• Vector Search: Suitable for unstructured text, it is the infrastructure for intelligent search;
• Large Model Application: Not only for text generation but also as an understanding and analysis tool, with great potential in the government affairs field;
• Open-Source Collaboration: Gather community wisdom to continuously improve algorithms and cover more data types.

The future development directions of CZR include:

• Real-time data stream processing to realize real-time contract analysis and early warning;
• Cross-language analysis to automatically process multilingual contracts and establish correlations;
• Predictive analysis to forecast procurement needs and price trends based on historical data;
• Visual presentation to display contract network relationships through interactive charts. This project provides an example for AI-enabled government transparency and is worth learning from by researchers and developers.