February 2024 Summaries

Haystack is an open source Python framework by deepset for building custom apps with LLMs. It allows developers to quickly try out the latest models in natural language processing (NLP) and integrates with various AI tools, including vector databases like DataStax Astra DB. In this tutorial, we walk through a simple example of using Haystack pipelines to store and retrieve data in Astra DB on a local machine. We demonstrate how to load data from the Star Wars Wikipedia page into the database, and then build a simple RAG application to answer questions about Star Wars using vector search.

Data preparation is a significant challenge for developers working on RAG (retrieval augmented generation) or generative AI applications due to the variety of difficult-to-use document types such as HTML, PDF, CSV, PNG, and more. Unstructured.io is a no-code platform that helps convert various document types into LLM-ready data and sets up GenAI data pipelines for transformation, cleaning, and generating embeddings for vector databases. The new integration between Unstructured.io and Datastax Astra DB enables developers to quickly convert common document types into vector data for highly relevant GenAI similarity searches. This integration allows users to build a simple but elegant RAG pipeline powered by an Astra DB integration that takes various data formats and uses Python code to create an LLM-based query engine, retrieving parsed data to provide insights to users. The process involves parsing documents using Unstructured, adding support for the Astra DB Destination Connector, setting up a RAG pipeline with Unstructured.io powered by Astra DB, and finally using LlamaIndex to connect to the newly created store and perform queries against it. This integration opens up the RAG and LLM world to challenging-to-parse documents, demonstrating the power of Unstructured.io and Astra DB together.

DataStax has launched Mission Control, a new tool designed to simplify the management of distributed databases in cloud-native applications, big data, and generative AI environments. The software is fully compatible with Cassandra and DataStax Enterprise, ensuring smooth integration with existing systems. It automates tedious processes and minimizes impact on running applications, allowing teams to focus on delivering business value rather than managing database intricacies. Mission Control is available at no cost for DataStax Enterprise users and can be added to Luna subscriptions for an additional fee.

DataStax is celebrating the three-year anniversary of its serverless database, Astra DB, which separates compute and storage for a massively scalable, cloud-native solution. The addition of vector search capabilities has made it easier to support generative AI applications. In a recent study commissioned by DataStax, GigaOm benchmarked Astra DB's performance and TCO against Pinecone's vector database, finding that Astra DB delivers superior performance and cost-efficiency with up to 80% lower TCO over three years. This significant cost advantage is due in part to the ability for applications to query data concurrently while ingesting and indexing without needing to pause queries to rebuild indexes.

The recent advancements in generative AI (GenAI) have significantly improved the way applications are built and interact with users. This article demonstrates how to use GenAI for car shopping by building a web application in Python, which utilizes DataStax Astra DB vector database and the CLIP model. The application allows users to search for cars using natural language or images, transcending traditional search methods that rely on user-provided text descriptions. This opens up new possibilities for more accurate and relevant search results across various industries.

Hornet, a leading social network and dating app for the global queer community, has integrated semantic vector search into its application to improve user profile matching. The company uses Apache Cassandra® and DataStax Astra DB databases in conjunction with OpenAI's vectorization technology. Storage Attached Indexes (SAIs) enable filtering on queries without specifying partitioning columns, making it easier to combine vector search with other criteria such as geographic distance or profile attributes. The Data API provides a simpler experience for developers and supports JSON document formats, including vector data types for semantic search combined with filtering. This integration allows Hornet to refine user matches based on distance from the user's address and other structured attributes, improving the overall user experience.