LAWCEDIC – AI RAG system
LAWCEDIC is an advanced AI-driven platform designed to assist users, such as legal professionals, students, and researchers, in legal research and information retrieval. The platform is associated with the Observatorio de Cibercrimen y Evidencia Digital en Investigaciones Criminales (OCEDIC), which is part of the law faculty at the Austral University in Argentina. LawCEDIC leverages artificial intelligence to provide semantic search, document retrieval, and contextual assistance, making legal research more efficient and accessible.
It’s built on a modern Retrieval-Augmented Generation (RAG) architecture, utilizing two distinct RAG systems: one dedicated to document loading and another for semantic search. The platform is deployed on Amazon Bedrock and uses a fine-tuned Llama 3.2 AI model for domain-specific legal tasks.
System Architecture and Process
LAWCEDIC’s built on a modern Retrieval-Augmented Generation (RAG) architecture, utilizing two distinct RAG systems: one dedicated to document loading and another for semantic search. The platform is deployed on Amazon Bedrock and uses a fine-tuned Llama 3.2 AI model for domain-specific legal tasks.
Key Technologies and Best Practices
- Vector Database: A scalable, low-latency vector database (e.g., Pinecone, Weaviate, Milvus, or Qdrant) is used to store and retrieve document embeddings, supporting hybrid search and metadata filtering for precise legal research
- RAG Architecture: The dual RAG system ensures robust document ingestion and high-precision semantic search, grounding AI responses in authoritative legal sources
- Amazon Bedrock: Provides managed infrastructure, model hosting, security, and compliance, as well as tools for prompt engineering, agentic workflows, and integration with external knowledge bases
- Fine-Tuned Llama 3.2: The model is adapted to the legal domain for improved accuracy, reduced hallucinations, and structured output generation
Task
Improve document insertion time by 50x and integrate semantic search through AI