Best of Vector Search — August 2024

1. 1
   Article

   MongoDB·2y

   MongoDB AI Course in Partnership with Andrew Ng and DeepLearning.AI

   MongoDB has partnered with Andrew Ng and DeepLearning.AI to launch a course focused on bridging database technology with modern AI applications. Titled 'Prompt Compression and Query Optimization,' this course covers Retrieval Augmented Generation (RAG) and uses MongoDB Atlas Vector Search to teach developers how to build efficient AI applications. The course also introduces techniques like prompt compression to reduce operational costs and showcases MongoDB's flexible document model ideal for AI data handling. It's tailored for developers interested in vector search, RAG applications, and AI optimization strategies.

   35
2. 2
   Article

   Community Picks·2y

   LLM app dev using AWS Bedrock and Langchain

   The post explains how to develop applications using Large Language Models (LLMs) with Amazon Bedrock and Langchain to perform tasks like Question Answering over large document corpora. It introduces the concept of retrieval-augmented generation (RAG), which uses document processing and vector embedding to fetch relevant document chunks for question answering. The process includes setting up LLM and embedding models, loading and splitting documents into chunks, creating a vector database using SingleStoreDB, and performing similarity searches to generate context-aware responses.

   21
3. 3
   Article

   Machine Learning News·2y

   Jina AI Introduced ‘Late Chunking’: A Simple AI Approach to Embed Short Chunks by Leveraging the Power of Long-Context Embedding Models

   Retrieval-augmented generation (RAG) involves breaking large documents into smaller text chunks for efficient information retrieval using embedding models. The release of jina-embeddings-v2-base-en, an open-source model with 8K context length, highlighted practical limitations in handling long documents. Late Chunking, a new approach, addresses these issues by applying the transformer layer to the whole text first, preserving contextual information and improving retrieval efficiency. Tests showed that late chunking outperforms traditional methods, especially for longer texts.

   17
4. 4
   Video

   Community Picks·2y

   Building Generative AI Applications in Go (continued)- Gari Singh, Google

   The post discusses building generative AI applications with Go, focusing on various frameworks and APIs provided by Google Cloud. Key highlights include the use of native Go APIs for AI models, the concept of prompt engineering, and orchestrating models using Lang Chain Go. It also covers retrieval augmented generation (RAG) and using vector embeddings for context, with practical demos involving tools like AMA and Gen Kit for better AI integration. The emphasis is on creating modular, extensible applications using these tools.

   16
5. 5
   Article

   Daily Dose of Data Science | Avi Chawla | Substack·2y

   The Utility of Vector Databases in LLMs

   Explore the surge in interest for vector databases, which are essential for storing and querying unstructured data. These databases are now crucial for enhancing the functionality of Large Language Models (LLMs) by embedding unstructured data into high-dimensional vectors. Learn how AssemblyAI's LeMUR framework simplifies building LLM apps on audio data, and how vector databases avoid the need for costly and complex fine-tuning of LLMs.

   16
6. 6
   Article

   MongoDB·2y

   Find Hidden Insights in Vector Databases: Semantic Clustering

   Vector databases, designed for efficient handling of large, multi-dimensional data, are increasingly crucial for generative AI applications. Semantic vector clustering within these databases can uncover hidden insights by organizing information meaningfully, improving text analysis, sentiment analysis, and knowledge classification. This technique can enhance decision-making, customer service, and employee satisfaction by identifying patterns and reducing data complexity. By leveraging MongoDB's secure and fully-managed vector database capabilities, organizations can unlock these hidden insights and boost overall business efficiency.

   13
7. 7
   Article

   Towards Data Science·2y

   Building a RAG Pipeline with MongoDB: Vector Search for Personalized Picks

   Explore building a personalized movie recommendation system using a Retrieval-Augmented Generation (RAG) pipeline and MongoDB’s vector search capabilities. By integrating large language models (LLMs) and structured data retrieval, this project showcases how to handle user queries and generate accurate recommendations. Steps covered include setting up the environment, data modeling with Pydantic, embedding generation via OpenAI's API, data ingestion into MongoDB, and performing vector search operations.

   12
8. 8
   Article

   Community Picks·2y

   Don’t use vector. Use halvec instead and save 50% of your storage cost

   Pgvector's latest release introduces quantization features like scalar and binary quantization, significantly enhancing storage efficiency and performance. The new half-precision 'halfvec' type reduces vector and index storage by 50% without compromising query performance. While scalar quantization shows impressive gains, binary quantization's recall with high-dimensional embeddings needs further exploration.

   12