Last Updated: November 21, 2025
Embedding Models
| Model | Use case |
|---|---|
OpenAI text-embedding-ada-002
|
Semantic similarity for text |
Sentence Transformers
|
Offline batch embedding |
CLIP
|
Image-to-text matching |
Similarity Metrics
| Metric | Behavior |
|---|---|
Cosine
|
Angle between vectors (scale-invariant) |
Dot product
|
Retains scale; use for dense retrieval |
Euclidean
|
Good for hierarchical spaces |
Query Flow
Encode the query, fetch candidates via ANN index, rerank with hybrid signals, and synthesize response.
💡 Pro Tip:
Normalize vectors before comparison when mix of scales could distort cosine distance.