Weaviate

What it is

Weaviate is a vector database that can handle embedding models directly. This means it can process and store data in a way that reduces the need for extra code to glue different components together.

It also supports a hybrid search approach, combining BM25 and vector search. This allows for more flexible querying of the data stored in the database.

Where it fits in a deployment stack

Vector databases like Weaviate are designed to solve the problem of efficiently storing and querying large amounts of complex data, such as vectors and embeddings. They typically sit alongside other components like data ingestion pipelines, machine learning models, and application servers.

In a production system, a vector database would normally take over the work of storing and querying the complex data, allowing other components to focus on their specific tasks. This can help simplify the overall architecture and reduce the amount of custom code needed.

By providing a centralized location for storing and querying complex data, a vector database can help improve the performance and scalability of the overall system.

How to evaluate it

To evaluate Weaviate, you need to consider how it will handle your specific use case and data. This involves testing its performance, querying capabilities, and integration with other components.

• Feed it a dataset with varying vector sizes and query it using different parameters to see how it performs.
• Inject failures into the system, such as network partitions or node failures, to see how it recovers.
• Inspect the output of the hybrid search queries to see how well it combines BM25 and vector search results.
• Measure the latency of the retrieval step under different concurrency levels to see how it scales.
• Compare the performance of Weaviate with and without the built-in embedding modules to see how much of a difference they make.
• Test its support for different data formats and schemas to see how flexible it is.

What to verify before you commit

Before committing to using Weaviate, you need to verify several aspects of the tool to ensure it meets your needs. This includes checking its licence, data handling, operational cost, failure behaviour, and potential lock-in.

• Check the licence terms to see if they allow for free use, modification, and distribution.
• Verify how it handles data privacy and security, including encryption and access controls.
• Estimate the operational cost of running Weaviate, including hardware, maintenance, and support costs.
• Test its failure behaviour to see how it responds to different types of failures and errors.
• Check for potential lock-in by seeing how easily you can migrate your data and applications to a different platform.
• Verify the level of community support and documentation available to help with any issues that arise.