July 2024 Summaries

Weaviate 1.26 introduces several new features and improvements, including improved range queries, async replication, Anthropic integration, bulk enable API-based modules, multi target vector search, tenant status changes, tenant offload to cloud storage, async Python client, scalar quantization, and a Korean tokenizer. These updates enhance data consistency, efficiency, and flexibility in Weaviate's search capabilities.

Weaviate, an AI platform, celebrates its fifth anniversary with over 10,000 developers using it for various projects. The platform's optionality in terms of machine learning models, frameworks, and deployment is a key reason for its popularity among users. Flexibility in architecture is also crucial as different use cases have varying requirements. Weaviate offers hot, warm, and cold storage tiers to help customers drive more efficient resource consumption. The platform has introduced an AI-native framework with modular GUI tools to accelerate developers' work. A Recommender service is now available in private beta, and new Weaviate Cloud Tools provide easier data management for both developers and non-technical users.

Retrieval-Augmented Generation (RAG) is a technique used in AI applications that involves integrating a comprehensive knowledge base into a retrieval system to enhance language model generation capabilities. This post explores techniques for improving every part of the RAG pipeline, including indexing, retrieval, and generation. Indexing methods discussed include simple chunking, semantic chunking, and language model-based chunking. Retrieval enhancement strategies involve hybrid search, query rewriting, and fine-tuning embedding models. Finally, generation improvements focus on autocut to remove irrelevant information, reranking retrieved objects, and fine-tuning the LLM on domain-specific data.

To celebrate the release of TypeScript v3 client, three events were held to showcase the new client to the JavaScript community and empower users building with it. The focus was on people using Angular, React, and Vue.js. Vector databases are becoming more common in AI-Native applications in the Python ecosystem and are inching closer to doing the same in the JavaScript ecosystem. Weaviate is committed to helping developers leverage embeddings for building experiences in products, AI-Native and beyond. The events highlighted the importance of search in applications, the value of multimodality, and the growth of fullstack (AI) engineers.

Retrieval Augmented Generation (RAG) is a design pattern that developers are implementing into their applications to leverage large language models with contextual data not part of the model's training dataset. Weaviate has built Verba, an example of RAG implementation, showcasing various chunking and retrieval techniques for building Retrieval Augmented Generation applications. This blog post explores how to run Verba locally with Weaviate Cloud or a local instance of Weaviate, and then connect it all up to Ollama for local inference. It also discusses alternative deployment methods for Weaviate, such as using Weaviate Cloud, Weaviate Enterprise Cloud, Dockerized Weaviate, and Kubernetes environment with Helm charts.

This guide provides a step-by-step approach to integrating Weaviate's generative AI capabilities with Apple's development environment for iOS and MacOS projects. It includes Python scripting to streamline Weaviate database creation, using native Python frameworks for data science tasks, and utilizing generative AI models within applications. The articles assume little or no knowledge of Python, generative AI, or Weaviate. Resources include working code in Swift, SwiftUI, and Python, as well as instructions on obtaining an OpenAI Key.