Overview

A retrieval-augmented generation backend built for a localization Q&A service, designed the way you'd run it in production: multiple tenants on one deployment, each with fully isolated retrieval, so one customer's documents never leak into another's answers. Embeddings run locally (no third-party embedding API), supporting multiple languages, and generation is grounded strictly in retrieved context.

Highlights

• Tenant-isolated retrieval on Django + PostgreSQL 18
• Local multilingual embeddings, no external embedding costs or data exposure
• Grounded generation for a localization Q&A use case
• Setup, database friction, and fixes documented in a companion Medium article

Stack: Django · PostgreSQL 18 · local embeddings · RAG

Links

• Code: github.com/saaim12/multilingual-multitenant-rag
• Write-up: Building a multilingual, multi-tenant RAG engine in Django