English lesson summaries

Persian RAG Academy is a free, project-based course on Retrieval-Augmented Generation for Persian text, written in Persian. Along the way learners rebuild parsrag, an offline-first Python library. These pages are short English summaries for international readers: what each lesson builds, the key ideas, and the measured results with the same numbers as the Persian lesson. They are not translations; every page links to the full Persian lesson.

About the numbers

Every number comes from the course's own scripts on a small, synthetic Persian corpus (25 documents, 66 original and 51 paraphrased questions) with offline stand-in models (HashingEmbedding, an extractive FakeLLM). They show how to measure, not how good a real model is. Real-model numbers are left for you to produce with the same tools; none were invented.

#	Lesson	Module	What it is about
00	Setup	parsrag.doctor	offline tooling, extras, grader
01	RAG in 50 lines	parsrag.naive_rag	the five steps and a baseline
02	Persian normalization	parsrag.normalize	Arabic letters, digits, ZWNJ
03	Loading documents	parsrag.loaders	broken PDF text and its repair
04	Chunking	parsrag.chunking	Persian sentences, heading context, tokens
05	Embeddings	parsrag.embeddings	passage-level benchmarks
06	Vector stores	parsrag.stores	in-memory, FAISS, pgvector
07	Hybrid search	parsrag.retrieval	BM25 + dense with RRF
08	Reranking	parsrag.rerank	retrieve many, rerank few
09	Grounded generation	parsrag.generate	citations and "I don't know"
10	Evaluation	parsrag.eval	answer metrics, bootstrap, CI gate
11	Agentic RAG	parsrag.agent	LangGraph agent and its cost
12	Production API	parsrag.api	FastAPI, streaming, caching
13	Capstone	parsrag.rag, parsrag.cli	PersianRAG, CLI, web UI
14	RAG security	parsrag.security	indirect prompt injection

Try the library

pip install "parsrag[loaders]"
parsrag init                       # offline preset; --preset ollama for local models
parsrag ingest docs/               # index a folder (saved; re-runs only embed what changed)
parsrag ask "هزینهٔ ارسال چقدر است؟"

The library reference, the CLI and the HTTP API are documented (in Persian) in the reference section; the code itself has English docstrings.