September 2024 Summaries

The text discusses measuring Retrieval Augmented Generation (RAG) apps and introduces the RAG Assessment (Ragas) framework, which consists of four primary metrics: faithfulness, answer relevancy, context precision, and context recall. These metrics help developers evaluate their GenAI apps by measuring performance rather than guessing. The text also provides a code example using LangChain, Redis, and OpenAI to create a simple RAG app for answering questions about financial documents. Additionally, the author explains how to generate test sets with the Ragas library and emphasizes the importance of creating challenging test sets to evaluate the performance of RAG apps accurately.

Redis 8, the latest version of the open-source in-memory data store, has been announced with seven new data structures and a faster query engine. The Community Edition now includes native support for JSON, time series, and five probabilistic types. This release enhances real-time database capabilities, offering advanced features like vector search, secondary indexing for full-text search, exact matching, geospatial queries, numeric data handling, and data processing. Redis 8 also improves session management, caching, and recommendation systems by enabling efficient search operations across active sessions and reducing bandwidth consumption. Additionally, the new version supports semantic caching through vector search capabilities, which are particularly useful for GenAI-assisted chatbots.

The langchain-redis package integrates Redis capabilities with the LangChain ecosystem, providing a powerful toolset for building fast AI apps, especially those powered by Retrieval Augmented Generation (RAG). This collaboration gives developers access to Redis features such as vector similarity capabilities, caching mechanisms, and chat history management. The package includes two caching mechanisms: RedisCache and RedisSemanticCache, which can improve the speed and cost-effectiveness of LLM interactions. Additionally, it provides a Redis-powered way to handle chat history, allowing for more natural and coherent conversations. With langchain-redis, developers can build AI apps that can handle complex, real-time interactions with ease, making it easier to create powerful AI applications.

Redis' Hash Field Expiration (HFE) feature offers improved performance, scalability, and resource efficiency compared to other solutions like TairHash and KeyDB. HFE enables active expiration of hash fields with minimal latency, allowing Redis to efficiently delete expired fields in the background. The ebuckets data structure, introduced as a potential replacement for the current EXPIRE implementation, provides significant benefits over traditional methods, including improved memory usage and support for adaptive expiration limits. Through extensive benchmarking, Redis demonstrates superior performance and scalability, making it an attractive choice for applications requiring efficient management of expiring fields within complex data types like hashes.

Redis has released its summer update on August 23rd, featuring new products and features aimed at helping developers build fast apps faster, including Redis for AI which addresses a common challenge in AI of getting into production faster with a package of tools. The Community Edition is now fully equipped with all the Redis Stack features, making it the most advanced and well-supported Redis to get started with. New data types like JSON, time series, probabilistic, and geospatial have been added, along with powerful search and query capabilities, including vector search in the free version of Redis. A new Visual Studio Code extension has also been released, allowing developers to easily build with Redis right in their IDE. Additionally, Redis Data Integration (RDI) has been announced, a tool designed to make apps faster by providing data federation without needing to redesign the entire data strategy. The Redis Cloud Essentials plan has also been launched on the Vercel Marketplace for Vercel users.