A collection is a named group of related objects in Weaviate, comparable to a table in a traditional database. It defines the fields stored with each object and the vectorizer configuration used to turn suitable text into embeddings. In a recipe-ingestion pipeline, a dedicated Airflow task ensures that the collection exists before recipes are stored. section 14.6 section 14.11
Creating a collection requires three coordinated choices:
- Choose a collection name, such as recipes, to identify the records logically. section 14.6
- Define properties as named fields with Weaviate data types, such as UUID for identifiers and TEXT for content. section 14.6.2
- Configure the vectorizer and embedding model that determine how vectorized text connects to the embedding service. section 14.6
Recipe collection schema
A recipe collection can contain recipe_uuid, recipe_name, chunk_uuid, and chunk. UUID identifiers can skip vectorization, while the recipe name and content chunk remain TEXT fields eligible for vectorization. The chunk stores the portion of recipe content used for retrieval. section 14.6.2
Collection creation should be repeatable without recreating an existing collection. Check the Weaviate instance first; if the name already exists, log the condition and return. Otherwise, create the collection with its vectorizer configuration and properties. Changing the schema requires deleting the existing collection before running creation again, so the dedicated Airflow task also provides a visible operational checkpoint. section 14.6.3
