Home

Welcome to the Legal-RAG Wiki

This wiki documents the design, architecture, and evolution of Legal-RAG.

It complements the source code by providing:

• architectural explanations and design rationale
• implementation notes and engineering trade-offs
• project roadmap and current status
• usage notes that are intentionally kept out of the README

Legal-RAG is an open-source engineering reference implementation for law-grounded, traceable Retrieval-Augmented Generation (RAG) systems, with a focus on Chinese Contract Law (《民法典·合同编》).

⚠️ Legal-RAG is not a legal advisory service.
It provides legal information assistance for research, education, and engineering purposes only.

---

What you’ll find here

This wiki will grow gradually and may include:

System Architecture

• End-to-end RAG pipeline overview
• Hybrid retrieval (FAISS + BM25)
• Routing logic and law_graph usage
• Offline vs. online data flow

Design Decisions

• Rationale behind architectural choices
• Why certain defaults are used
• Prompt constraints and safety considerations
• Trade-offs between simplicity, correctness, and extensibility

Roadmap & Status

• Implemented features
• Planned extensions
• Explicit non-goals (what the project intentionally does not do)

Advanced Usage Notes

• Indexing strategies and metadata design
• PDF ingestion and incremental indexing
• Conversation-aware retrieval (when applicable)

---

How to use this wiki

• New readers: start with the architecture overview.
• Code readers: refer to design notes to understand why the code is structured this way.
• Questions or ideas: please use GitHub Discussions.
• Bugs or actionable tasks: please use GitHub Issues.

---