July 2024 Summaries

Qdrant 1.10's new Query API revolutionizes the process of building hybrid search systems by allowing users to combine various search methods directly on the server side, eliminating the need for additional services. This update introduces capabilities for creating complex search pipelines, including nested multistage queries, and supports multi-vector representations, which enhance retrieval quality by capturing text nuances. The API facilitates the fusion and reranking approaches, enabling efficient and flexible search experiences that can be tailored to specific use cases using standard evaluation metrics like precision@k and NDCG. The system can handle both sparse and dense vectors, offering options such as Matryoshka embeddings and late interaction models, while the Query API's versatility empowers users to experiment with different setups for optimal results. By integrating these advanced features, Qdrant aims to provide scalable and robust search solutions, as illustrated in their recent webinar on constructing hybrid search systems.

Kairoswealth, a comprehensive wealth management platform, has transformed its operations by integrating Qdrant, a scalable vector database, to enhance its AI-driven insights and service offerings. Faced with performance, scalability, and memory inefficiency issues using a previous vector database, Kairoswealth opted for Qdrant due to its high performance, memory efficiency, and robust support. Qdrant efficiently manages large datasets, enabling rapid similarity queries and improved recommendation systems, while its compact storage reduces infrastructure costs. This partnership allows Kairoswealth to innovate within the wealth management sector, notably through the development of the Kairos Terminal, which leverages Generative AI to automate tasks and expand service capabilities. The integration of AI technologies aims to automate back-office operations and scale services to a wider client base, positioning Kairoswealth as a leader in technologically advanced wealth management solutions.

BM42 is introduced as an evolution in lexical search, aiming to combine the strengths of the traditional BM25 algorithm with the capabilities of modern transformer models to address the limitations of existing search systems like SPLADE. While BM25 has been a cornerstone in search algorithms due to its effective term importance calculation, the rise of dense embeddings and hybrid search systems highlighted its limitations, particularly in scenarios like Retrieval-Augmented Generation (RAG), where document lengths are shorter. BM42 retains the inverse document frequency (IDF) component of BM25 but introduces a novel approach to term importance using transformer-derived attention matrices, allowing for more semantic relevance in document scoring. It addresses tokenization challenges by retokenizing after attention weight extraction, thereby enhancing accuracy without extensive re-training. Although BM42 demonstrates improvements in areas like query inference speed and memory footprint, it acknowledges the continued relevance of BM25 for larger documents and emphasizes the benefits of hybrid models that leverage both sparse and dense embeddings to optimize search results across various contexts.

Qdrant 1.10 introduces significant advancements, including a Universal Query API that consolidates all search APIs into a single endpoint, enhancing efficiency by reducing separate search requests to a single parameterized query. This version also incorporates the Inverse Document Frequency (IDF) mechanism into its core, enabling more dynamic and automatic updates for document frequency calculations, and introduces BM42, which combines IDF with transformer-based attention matrices to improve text retrieval. The update also supports multivector search compatible with models like ColBERT, enhancing semantic understanding and retrieval quality, and introduces a new float16 datatype for vectors, significantly reducing memory usage and improving processing speed. Additionally, Qdrant's Rust client has been revamped for easier use and safe concurrent execution, and the system now supports S3 snapshot storage for cost-effective and secure data management. The Issues API is also introduced to help users manage performance issues and misconfigurations efficiently.