Embeddings Primer
How Meridian turns recipes into vectors you can search, cluster, and compare.
What is an embedding?
An embedding is a list of floating-point numbers — typically 768 or 1536 of them — that represents the meaning of a piece of text. Two recipes with similar embeddings are semantically close: same cuisine, overlapping ingredients, or analogous techniques.
How Meridian generates them
Every recipe you save passes through a transformer model. We extract the title, ingredient list, and instructions, concatenate them into a single prompt, and run inference. The output vector is stored alongside your recipe metadata.
Example
"Thai green curry with coconut milk, basil, and jasmine rice" → [0.023, -0.841, 0.117, …, 0.632]
What you can do with them
- Semantic search — find "comfort food" even when the recipe never uses those words.
- Deduplication — detect near-duplicate recipes across your collection.
- Clustering — auto-group recipes by cuisine or flavor profile without manual tagging.
- Recommendations — surface recipes similar to ones you already love.
Storage and cost
Embeddings are compact. A 1536-dimensional vector is about 6 KB. Even 10,000 recipes consume under 60 MB. Meridian stores vectors in a purpose-built index so similarity queries return in milliseconds.
Ready to try it? Recipe search →