Document search
Find documents from concepts and questions. Filter by collection, date or permissions to retrieve results within the query’s scope.
Use cases / Data and knowledge
Semantic search
Build semantic search that understands what people want to find.
Help users find documents, products and content even when they use different words. Semantic search uses embeddings to connect the intent of a query with the meaning of your data.
With QDivZero Retrieval, you can build vector search for a catalog or knowledge base, combine similarity with filters and reuse the index for recommendations or RAG applications.
Search with context
The same idea can be written in many ways.
Find content by meaning, even when queries use different words. Combine similarity with your application’s filters.
Product search
Match customer needs to catalogue listings. Combine semantic relevance with categories, attributes and availability.
Knowledge bases
Retrieve articles and answers from your knowledge base. Help users find content even when they do not know its exact terminology.
Similarity recommendations
Find related documents or products using embeddings. Use metadata and business rules to prepare relevant recommendations.
Build it with QDivZero
Prepare the index. Search by meaning.
Represent content and queries with the same embedding model. Retrieve results by similarity, apply filters and add reranking if needed; your application presents the documents or products found.
Prepare your data
Content
Prepare documents, products, or items to search.
Embeddings
Generate vectors representing each collection item.
Vector index
Associate vectors with original data and metadata.
For each query
Query
The user describes what they want to find.
Embedding
Represent the query using the index embedding model.
Search
Compare vectors and apply collection filters.
Results
Return content and metadata to your application.
Open-weight / Hugging Face
You can start with…
The embedding model determines how queries and content are represented. Compare results with your application’s languages and data; add reranking when you need to order retrieved candidates more precisely. Response time includes embedding generation and reranking when used. Compare the complete process with your collection size and expected search concurrency.
Embeddings
Qwen3-Embedding-8B
Represent documents and queries as vectors for semantic similarity search.
View model on Hugging FaceReranking
Qwen3-Reranker-8B
Reorder retrieved results according to their relevance to the query.
View model on Hugging FaceSemantic search
Semantic search: frequently asked questions
What is AI semantic search?
Semantic search finds content by meaning using embeddings and vector similarity. It can retrieve documents or products related to a query even when descriptions use different words. Apply it to your own data with QDivZero Retrieval.
How do I build semantic search for documents?
Embed the content, store vectors in an index, and keep document metadata. Embed each incoming question using the same model and search for similar vectors. Your application displays results and references.
How does semantic search differ from keyword search?
Keyword search finds text matches. Semantic search compares meaning represented in vectors, allowing queries and content phrased differently to match. Choose the approach based on the information your users need to find.
Can I use semantic search in a product catalog?
Yes. Embed catalog descriptions and characteristics to match them with search intent. Combine similarity and metadata filters to narrow results by category or other available product attributes.
Which embedding model should I choose for vector search?
Choose a model suited to your languages, content type, and document length. Qwen3-Embedding is a starting point for comparing results. Use the same compatible model for indexing content and representing queries.
Can I reuse semantic search for RAG or recommendations?
Yes. A vector index can retrieve context for RAG or find similar items for recommendations. Your application determines how to use results. Changing the embedding model usually requires regenerating vectors to keep the index compatible.
How do I improve semantic search relevance?
Evaluate real queries, improve content descriptions and use filters to narrow the collection. Passage size and the embedding model affect the candidates retrieved. You can add a reranker to reorder them and measure whether useful results appear earlier.
How much does vector search cost to implement?
It depends on indexed items, embedding generation and query volume. Index updates and reranking also contribute to the workload. QDivZero pricing helps you plan capacity; compare performance with a sample of your catalog or documentation before expanding the search.
Ready to help your search understand each query?
Build semantic search for documents, products and content with your own data. Deploy an embedding model, prepare an index in QDivZero Retrieval and connect the results with your application experience.