A domain-agnostic, multi-tenant RAG knowledge platform. The same pipeline — PDF ingestion, chunking, embeddings, pgvector similarity search, and citation-grounded answers — serves legal documents, company policies, research papers, compliance manuals, and internal knowledge bases. The architecture stays identical across industries; only the content changes.
Context: We needed vector search that respects tenant boundaries without operating an extra service.
Decision: Store embeddings in PostgreSQL with pgvector alongside relational data, keeping the stack simple, transactional, and tenant-aware with trivial joins.
Context: The original concept was law-firm specific, but the underlying pipeline is identical for any document corpus.
Decision: Present the system as a reusable multi-tenant RAG platform, demonstrating versatility across industries while keeping a single technical architecture.