YAML Metadata Warning: empty or missing yaml metadata in repo card
Check out the documentation for more information.
license: mit datasets:
- Pinizov/credit-guardian-laws-bg language:
- bg metrics:
- accuracy base_model:
- pytorch/gemma-3-27b-it-FP8 tags:
- finance
- legal
- agent
Credit Guardian – Bulgarian Legal Data Agent 🇧🇬
Overview
Credit Guardian is a Bulgarian legal-data and credit-law assistant built on top of Gemma 3 27B, optimized for extracting, organizing and querying legal texts related to consumer and mortgage credit.
The system is designed for tasks like detecting potentially illegal or unfair clauses in loan contracts, exploring credit-related legislation, and supporting downstream RAG-based analysis in Bulgarian.
Data
The core knowledge base is stored in services/import_laws/data/laws.csv, containing tens of thousands of full legal articles in Bulgarian (not summaries).
Legal content is sourced from EU credit directives, Bulgarian legislation portals, central bank regulations, and curated analyses of real contracts, while respecting the terms of use of each source and without redistributing proprietary texts.
Agents and architecture
The system is organized as a set of cooperating agents orchestrated by a master controller:
- Master coordinator –
master_extraction_agent.pyruns the full pipeline, orchestrates all sources and updates the local database with new or changed legal texts. - EU scraper –
comprehensive_eu_credit_scraper.pycollects EU directives and related documents (e.g., consumer credit, mortgage credit, unfair terms) from official EUR-Lex pages in Bulgarian and normalizes them into article-level records. - RAG search agent –
rag_search_agent.pyqueries Bulgarian legal portals (such as commercial legal databases, central bank sites and EU portals) and ingests relevant laws, regulations and consumer-protection materials. - Supabase extractor –
extract_supabase_info.pyloads preprocessed analyses from a Supabase database, including examples of unlawful or problematic clauses observed in real-world credit agreements.
After extraction, the consolidated corpus can be indexed in a vector database (for example Qdrant) to power retrieval-augmented generation and contract analysis workflows.
Usage
Running the full pipeline
From your project root datasets: - Pinizov/credit-guardian-laws-bg