Sanity Context and content APIs (opens in a new tab)
The official reference for the tools used in this lesson.
Link
Metadata and filtered retrieval
Metadata turns semantic search into scoped semantic search. Filtering by source, type, or date removes whole classes of wrong answer.
Lesson 2.212m 24sAdvanced7.2k students
Chapters
12:24- 0:00Metadata problem with RAG
- 0:45Using Lang Extract for Metadata Extraction
- 1:46Building the Retrieval Augmented Generation System
- 2:22Setting Up the Environment and Sample Data
- 3:27Creating the Metadata Extraction Pipeline
- 6:56Implementing Metadata Filters in Vector Store
- 8:51Running the Example and Viewing Results
What this lesson covers
- Attach source, type, and timestamp to every chunk
- Filter before ranking to cut the candidate set
- Use metadata to attribute answers back to a source
This lesson sits in Ingestion and Indexing, part of Retrieval-Augmented Generation from Scratch. It assumes what came before it and leads directly into the next lesson in the module.
In this lesson you will
- Attach source, type, and timestamp to every chunk
- Filter before ranking to cut the candidate set
- Use metadata to attribute answers back to a source
Resources
Personal notes are coming soon.