Draft answers from approved sources.
Build a small source retrieval flow with access checks, citations, and a no-answer path.
Can the assistant show the source for an answer without exposing restricted information?
A fictional workshop has public opening information and a private staff note. Build an extractive answer prototype for a public visitor. The result stays a draft. Record when no accessible current source supports the question.
Build it, step by step
- 01
Load the documents and preserve id, audience, and source version.
- 02
Filter access before searching or constructing model input.
- 03
Implement a simple lexical retriever and document its limitations.
- 04
Return an exact source excerpt with its id or an explicit review result.
- 05
Test missing answers, stale versions, restricted sources, and a source that contains hostile instructions.
Your project checklist
- Restricted text never enters a public answer context.
- Each draft includes an allowed source id and excerpt.
- No evidence produces a review result.
- Source versions and freshness rules are recorded.
- The report explains that literal support does not prove complete factual correctness.
Free, open and run by a nonprofit.
This page is part of AI Automation, a free course of 20 modules and 60 lessons in the engineering.fit( ) Learning Lab. The Learning Lab is built and run by engineering.fit( ), a nonprofit founded by Abdullateef Opeyemi Bakare that helps African engineering students gain research experience. You do not need an account or pay a fee to learn.