Building A Legal RAG App in 36 Hours

A walkthrough of building a production-ready legal contract RAG application in 36 hours using Weaviate's Query Agent and Agent Skills library. The architecture uses ColQwen multivector embeddings to encode PDF pages as visual tokens (avoiding OCR/chunking), Muvera compression for efficiency, and three separate Weaviate collections for different contract types. The Query Agent handles agentic search by inspecting schema, constructing structured queries with filters, reranking results, and synthesizing grounded answers. The app supports two modes: Search Mode for discovery and Ask Mode for direct Q&A with cited sources. Setup is demonstrated via a single prompt in Claude Code or Cursor using the CUAD public dataset.