GenAI at the Presidential Academy
A new solution from coMind for legislative analysis
The Presidential Academy, backed by extensive expertise and years of experience in the legal field, set out to strengthen its proficiency in GenAI. The goal was to develop an innovative solution that would enable efficient interaction with legal information and promptly find answers to relevant questions about legislation and regulatory frameworks. The Academy carried out the project in partnership with coMind, a specialist in large language model (LLM) technologies.
- 1+ million
Active regulations in federal legislation - 2,500+
Annual legislative updates - 800+
Federal laws
Ekaterina Tarasova
Project Director, Graduate School of Public Administration, Presidential Academy
Leveraging GenAI for legislative analysis is a leap forward in automating legal research. Instead of spending hours studying documents, users can now simply ask a question in natural language and receive an accurate answer
Project Implementation
From Analysis to Deployment
The coMind team began with an in-depth study:
- Structure of Legal Databases: Examining how legislative acts, legal norms, and expert commentary are interconnected.
- User Query Scenarios: Identifying common requests—from simple searches for legal provisions to complex legal interpretations.
- Contextual Analysis: Designing algorithms that account for nuances in user questions to avoid ambiguity in responses.
Building on this analysis, the team developed an architecture that ensures:
- High Accuracy of Responses: Utilizing advanced text-processing algorithms based on LLMs.
- Fast Document Indexing: Enabling instant access to the required information.
- Flexible Data Updates: Continuously integrating frequent legislative changes to keep information current.
Challenges and Solutions
Key aspects
- Integration with GigaChat by Sber: For natural language processing and improved user interaction.
- Development of a Specialized Search Engine: Taking into account the specifics of legal queries to enhance result relevance.
- Multi-Level Answer Validation: Ensuring the reliability of the information provided.
Challenges and Solutions
Ambiguous Legal Terminology
Ambiguous Legal Terminology
- Challenge: Many legal terms carry different meanings in various contexts.
- Solution: Creation of a semantic analysis mechanism to pinpoint the exact meaning of terms in each context.
- Challenge: Extensive information arrays, including laws, by-laws, decrees, and case law.
- Solution: Development of high-performance indexes and deployment of cutting-edge Natural Language Processing (NLP) technologies.
- Challenge: The need to provide current and precise information.
- Solution: Implementation of a source-ranking system that prioritizes official publications.
Results and Future Development
- up to 40%
Reduction in workload for legal departments - 100+ hours
Saved per month on legal queries - over 80%
Accuracy of responses thanks to multi-level validation
Collaboration between the Presidential Academy and coMind has led to the creation of a unique product—the “Vector Extractor” of Regulatory and Legal Documentation. This solution sets new standards in efficiency and accuracy for legislative information searches.
Plans for the Future:
Plans for the Future:
- Expanded Functionality: Integrating additional databases and automatically updating legal provisions.
- Corporate Market Entry: Offering the solution to large enterprises that need prompt legal expertise.
- Voice Interface Development: Introducing voice assistants for more convenient system interaction.