AI Search
Retrieval that ranks by meaning rather than by keyword match.
Overview
AI search replaces term matching with embedding similarity, optionally reranked, so a query finds documents that mean the same thing without sharing words. It solves the vocabulary mismatch problem that has always limited keyword search: the page titled ending your subscription should be found by someone searching how do I cancel. It exists because embedding models became cheap enough to index an entire corpus, and it is the retrieval half of every question-answering system.
Technology Graph
How AI Search connects to the rest of the stack.
Uses and depends on
- Embedding concept Techniques AI Search uses Embedding High ✓ source
- Reranking technique Techniques AI Search uses Reranking High ✓ source
Applied to
- BEIR benchmark Benchmarks BEIR used for AI Search High ✓ source
- Bi-Encoder concept Techniques Bi-Encoder used for AI Search High ✓ source
- CLIP model Multimodal CLIP used for AI Search High ✓ source
- Cohere Embed model Embedding Models Cohere Embed used for AI Search High ✓ source
- E5 model Embedding Models E5 used for AI Search High ✓ source
- Elasticsearch database Search Engines Elasticsearch used for AI Search High ✓ source
- FAISS library Vector Databases FAISS used for AI Search High ✓ source
- HNSW algorithm Algorithms HNSW used for AI Search High ✓ source
- Hybrid Search technique Techniques Hybrid Search used for AI Search High ✓ source
- Jina Embeddings model Embedding Models Jina Embeddings used for AI Search Medium ✓ source
- MS MARCO dataset Datasets MS MARCO used for AI Search High ✓ source
- PostgreSQL database Databases PostgreSQL used for AI Search High ✓ source
- RAG technique Techniques RAG used for AI Search High ✓ source
- Redis database Databases Redis used for AI Search High ✓ source
- SQLite database Databases SQLite used for AI Search High ✓ source
- Typesense database Search Engines Typesense used for AI Search High ✓ source
- Vespa database Search Engines Vespa used for AI Search High ✓ source