RAG Formation · Reference

One question through the RAG system

A standalone cheat sheet for learning and testing RAG quality on this site.

$ reference --status
▸ course RAG Formation
▸ mode standalone reference
▸ status Complete
● learn → test → verify

Reference sheet

The guided RAG Formation course content is complete; this reference is available for self-directed practice.

Core path

HTTP request → route → service → retrieve → assemble → cite → generate → response

Stage	File / function	Responsibility
1. HTTP entry	`routes/qa.py` `ask_question()`	Receives `QuestionRequest` and passes it to the QA service.
2. Orchestration	`qa_service.py` `answer_question_internal()`	Checks the input, resolves services and configuration, and coordinates the request.
3. Retrieval	`search.py`	Searches Qdrant chunks, Qdrant summaries, Meilisearch, and LightRAG.
4. Context	`utils.py` `build_context()`	Splits the budget across channels and keeps the items that reach the model.
5. Citations	`utils.py` `sources_from_items()`	Builds the source list from evidence that survived context assembly.
6. Prompt	`utils.py` `build_prompt()`	Tells the model to use the context, cite sources, and refuse unsupported answers.
7. Generation	`qa_service.py` `completion.generate()`	Generates the final answer from the question and assembled context.
8. Response	`models.py` `AnswerResponse`	Sends back the answer, sources, and diagnostic stats.

Before generation: check `/api/v1/search` and `context_items_included`.
At citation time: compare `sources` with the items that actually made it into the context.
After generation: compare the answer with both the evidence and the expected answer.
For latency: inspect all four retrieval calls—especially LightRAG—and the final completion call.

Continue learning: Open the matching standalone lesson · Course roadmap