RAG system for legal case-file analysis

Legal proceedings generate large volumes of heterogeneous information. Relevant statements are spread across pleadings, emails, chat histories, evidence, and other documents. Purely manual search is time-consuming and makes it difficult to trace relationships consistently across the complete case.

At the same time, every answer must remain verifiable. An AI assistant must not return isolated assertions; it needs to show which documents and passages support its response.

The RAG system ingests the provided case documents, extracts their content, and splits it into semantically meaningful sections. These sections are indexed and retrieved as targeted context whenever a question is asked.

Documents and information can be organised into evidence, opposing statements, and client statements. The law firm can ask specific questions as well as broader questions about contradictions, timelines, and the available evidence.

The web interface was built with Next.js and TypeScript. A Python backend using FastAPI handles document processing, retrieval, and controlled communication with the language model.

PostgreSQL stores cases, documents, text sections, and associated metadata. Clear system boundaries allow model operation, hosting, and storage to be adapted to the confidentiality requirements of the law firm.

The solution makes case-specific knowledge available through a natural-language interface and connects relevant answers to traceable sources. Extensive files can be explored more quickly while important relationships remain systematically verifiable.