CogDoc
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
π CogDoc LLM: Privacy-Preserving Document Intelligence CogDoc is an offline, privacy-first legal and policy document analysis system. It utilizes a multi-model pipeline to segment clauses, detect contradictions (NLI), and provide AI-driven risk assessments without sending data to the cloud. π Features Offline Execution: Runs entirely on local hardware for data sovereignty. Multi-Model Pipeline: * Mistral-7B (GGUF): For high-level reasoning and risk assessment. DeBERTa-v3-small: For Natural Language Inference (NLI) and clause cross-checking. spaCy: For high-speed document segmentation. Interactive Dashboard: Built with React and FastAPI. π Datasets This project utilizes multiple benchmark datasets for legal and policy document analysis. These datasets cover tasks such as clause classification, contract understanding, legal entailment, and summarization. πΉ LEDGAR A large-scale dataset for contract clause classification derived from SEC filings. π https://github.com/IBM/ledgar πΉ CUAD (Contract Understanding Atticus Dataset) A benchmark dataset for extracting and classifying important clauses in legal contracts. π https://github.com/TheAtticusProject/cuad πΉ ContractNLI A natural language inference dataset for legal contracts, enabling reasoning over contractual clauses. π https://stanfordnlp.github.io/contract-nli/ πΉ BillSum A dataset for summarization of U.S. Congressional and California state bills. π https://github.com/FiscalNote/BillSum πΉ ILDC (Indian Legal Documents Corpus) A dataset focused on Indian legal case documents for classification and judgment prediction tasks. π https://github.com/Legal-NLP-EkStep/ILDC β οΈ Usage Note All datasets are publicly available. Please refer to the respective repositories for licensing terms, access conditions, and citation requirements. π Citation If you use these datasets, please cite the original sources as specified in their respective repositories. π οΈ Setup & Installation 1. Clone the Repository PowerShell git clone cd legal-ai-app/app 2. Create and Activate Virtual Environment It is recommended to use a dedicated virtual environment to avoid dependency conflicts. PowerShell # Create the environment python -m venv backend-venv # Activate it (Windows) .\backend-venv\Scripts\activate 3. Install Requirements Install the core dependencies and fix the numpy version for compatibility with spaCy. PowerShell pip install -r requirements.txt pip install "numpy
Fields of Study & Subjects
Artificial Intelligence in Law
Topic Modeling
Handwritten Text Recognition Techniques
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 26 March 2026
- Language
- EN