December 2023 Summaries

The December 2023 Vespa Newsletter highlights several advancements and updates in the Vespa platform, including the introduction of a global-phase cross-hit normalization feature, which allows for more refined ranking by evaluating global lists of hits on container nodes. This update includes new normalizing functions like normalize_linear, reciprocal_rank, and reciprocal_rank_fusion, aimed at providing better control over ranking factors. Pyvespa, the platform's API, now offers improved data feeding capabilities and supports global-phase ranking expressions, along with streaming mode options. Vespa has also enhanced security with Token Authentication for data plane access, eliminating the need for certificates. Additional features include the ability to download active application packages for easier replication and the unpack_bits function for converting int8 input to float values, akin to numpy.unpackbits. The newsletter also celebrates the successful year Vespa has had in the search and recommendation industry, introduces a new visual identity, and encourages participation in the Advent of Tensors challenge, while offering a glimpse into recent blog posts and the opportunity to try Vespa Cloud for free.

This blog post provides a comprehensive guide on integrating LangChain with Vespa streaming mode to create cost-efficient RAG (Retrieval-Augmented Generation) applications over sharded data. It explains how Vespa’s streaming search solution allows for efficient data grouping by integrating a sharding key into the Vespa document ID, enabling low-latency searches without using memory, which significantly reduces deployment costs. The article details a step-by-step process of deploying a Vespa application using PyVespa, processing PDFs with LangChain, and developing a custom LangChain retriever that utilizes Vespa's capabilities to extract meaningful context from PDF documents. Additionally, it demonstrates the deployment to Vespa Cloud and the querying of data using a custom retriever, emphasizing the benefits of Vespa's streaming mode, such as eliminating precision compromises and achieving higher write throughput without the need for index builds.

Vespa.ai, recently established as an independent company, has unveiled a new logo and visual identity to mark its progression into a new era, emphasizing its bold, Scandinavian roots. The newly introduced geometric logo is designed to symbolize Vespa's no-nonsense, high-performance ethos and can be interpreted as a versatile container of data, aligning with the concept of "Moving mountains, not data." The updated brand identity includes a refreshed typographic palette and a color scheme inspired by Scandinavian natural elements, enhancing cross-platform legibility and maintaining a professional tone. The high-contrast brand icons and photographic style further reflect Vespa's origin, highlighting its solid, agile, and complex nature. This change is part of a broader update, with new features for the Vespa platform and Vespa Cloud anticipated in the coming months.