beginner pathway
Local AI, Evidence Search, and Introductory RAG
Learn how grounded answers differ from unsupported model guesses.
Download course manifest
Evidence First, Then Answer
A grounded answer is built from inspectable evidence. The source can still be wrong, incomplete, or outdated, so the answer should show citations and uncertainty instead of hiding the basis.
Exercise
Write one question that requires a source and one that is just a personal preference.
Assessment
What should an answer do when it lacks evidence?
The honest response is to say evidence is missing or uncertainty remains.
Sources
Simple Retrieval for Source Finding
Retrieval starts with finding candidate passages. Lexical search matches words directly. Hybrid retrieval can add embeddings and reranking, but access control and source review still matter.
Exercise
Search this site for read_text and name the course and lesson that mention it.
Assessment
Is a search result automatically proof?
Search results are leads. Inspect the source before relying on the claim.
Sources
Mini Project: Cited Answer Card
Retrieval-augmented generation is useful only when users can inspect what was retrieved and understand what the model added. A beginner can start with a deterministic cited answer card before using any model.
Exercise
Use Ask AI on this site for a course question, then export or copy the cited answer summary.
Assessment
What is one private boundary for public Ask AI?
Public interfaces must not expose engineering tools or another user's private data.
Sources
Final mini-project
Create a cited answer card from local course evidence and explain its uncertainty.