RAG in the browser

Unit 1 dari 6Cutting text

See the finished app

See the finished app first

You will build: an app that reads the Toko Batik Laras SOP (or your own PDF) and answers buyers' questions with page citations.

Before learning anything, take a peek at the app. On the right is a preview: one PT Sinar Maju employee question about annual leave answered by four versions of an HR assistant app. Click Next to see each version, and change the question if you're curious. Each version changes just one thing.

Then answer the question under the preview.

What's inside, one line per unit

Unit 1: Cutting textCut a long company regulation into tidy chunks.
Unit 2: Finding similar chunksFind the regulation chunk closest in meaning to an employee query.
Unit 3: From document to indexBuild an index: article chunk, page number, and its vector.
Unit 4: Asking with citationsAnswer with citations you can check against the actual article.
Unit 5: Checking data with SQLCheck article completeness with SQL before using them.
Unit 6: Final project: your RAG appFinal project: a RAG app that answers from your own regulation PDF.

Next up

Before you start: Before you start: Python in 5 minutes

What gets checked

In which version can the HR officer check the answer against the original regulation clause?

Zen VectorThis is how your app will work

Change the question

Version 1 of 4 · No document

What changed: Starting point: the model gets no document at all.

Context sent to the model

(none)

Model answer

Usually 7 days.

Without a document the model only guesses. It sounds sure, but it's wrong.

In which version can the HR officer check the answer against the original regulation clause?