July 2023 Summaries

Healthsearch is an open source demo that performs semantic and generative search on supplement reviews, retrieving the most related products for specific health effects based on user queries. It uses large language models (LLMs) like GPT4 to translate everyday language queries into structured query format called GraphQL query, which can extract information about filters, sorting, and limits directly from the context of the query. The demo also decodes reviews with semantic search and applies generative search to enhance user experience by generating product summaries based on user queries. Healthsearch is designed for intuitive search and can be adapted to any other domain.

This article discusses how Test-Driven Development (TDD) can be made easier with Embedded Weaviate, a feature that allows developers to instantiate Weaviate directly from the client. The author provides examples of integration tests for applications using Weaviate and offers tips on scoping tests, testing with embedded Weaviate, and setting up continuous integration/continuous deployment (CI/CD) pipelines. By incorporating TDD and utilizing tools like Embedded Weaviere, developers can improve the robustness and reliability of their applications while ensuring that they are always in a deployable state.

Weaviate 1.20 introduces several new features including multi-tenancy, PQ + Re-scoring, Autocut, search re-ranking, a new hybrid search ranking algorithm, and more. Multi-tenancy allows for efficient storage of data from millions of tenants in one setup. PQ + Re-scoring improves vector search speed with minimal impact on accuracy. Autocut enables users to set a threshold for the number of result groups retrieved. Search re-ranking introduces rerankers for improved final results sets, and a new hybrid search ranking algorithm is introduced for more nuanced ranking of objects.