Connect a vector database

Vector databases and UltiHash

A vector database stores embeddings - numerical representations of data that capture meaning and context. Unlike traditional databases, vector databases enable similarity-based searches, returning results that are semantically related rather than relying solely on exact keyword matches. Common use cases include:

• Semantic search (text, images, multimedia)

• Recommendation engines

• Retrieval-augmented generation (RAG)

How you can connect it to UltiHash

In most use cases, the vector database is only responsible for storing and indexing embeddings. The actual raw data (such as documents, images, or videos) is stored separately in UltiHash.

To integrate UltiHash with a vector database, you'll typically use a middleware application. This middleware will:

• Perform semantic search via the vector database to get references (IDs or filenames) of relevant data.

• Query UltiHash using those references to retrieve the actual raw data.

• Return or serve the retrieved data to the end user or consuming application.

UltiHash itself does not directly communicate with the vector database - the middleware application orchestrates the integration.

Example pipeline

An example of a pipeline using this structure looks like:

• Store your raw data (e.g., images) in UltiHash.

• Generate embeddings from this raw data and store these embeddings in your chosen vector database.

• A middleware application receives a user query, performs a vector search against the database, and obtains references to matching items.

• The middleware then retrieves the actual files from UltiHash based on those references.

• Finally, the middleware serves the raw data (such as images or documents) to the user.

For a detailed example of this integration in action, please see our full case study using vector database Zilliz (Milvus) to set up a simple RAG pipleine: Why does vector search need object storage as its foundation?