Building Production RAG, One Arabic Tutorial at a Time

What it does mini-rag is a minimal question-answering application built to demonstrate Retrieval-Augmented Generation. The repository serves as the companion codebase for a 25-part Arabic YouTube course, with each episode mapped to a git branch so learners can follow the progression from an empty FastAPI skeleton to a deployed system.

The interesting bit Most tutorial repos dump a finished product and call it a day. This one evolves in public: the course migrates the database from MongoDB to Postgres mid-stream, adds PgVector for semantic search, and wires up Celery workers, Grafana, and Prometheus as if you were shipping to real infrastructure. The instruction is entirely in Arabic, making it a rare full-stack GenAI resource for that audience.

Key highlights

• 25 Arabic video episodes with matching git branches (tut-001 through tut-017).
• Full production stack: FastAPI, SQLAlchemy/Alembic, Postgres, PgVector, Celery, Flower, Docker.
• Supports both cloud LLMs (OpenAI) and local inference via Ollama.
• Includes monitoring (Grafana, Prometheus) and a Postman collection for API testing.

Caveats

• The educational content—video narration and explanations—is in Arabic only.
• Several consecutive videos share the same branch (e.g., tut-008 covers four episodes), so the code deltas between some lessons are not isolated.
• The README is essentially a quickstart guide and video index; architectural decisions are explained in the videos, not in text.

Verdict Arabic-speaking developers who want a guided, opinionated tour from “hello FastAPI” to “deployed RAG with workers” will get the most value. If you are looking for a drop-in RAG library or English-language documentation, this is not your repo.