Neum AI Project Overview
Neum AI is an innovative data platform designed to assist developers in utilizing their data to enhance Large Language Models (LLMs) through Retrieval Augmented Generation (RAG). This revolutionary platform is adept at extracting data from existing sources such as document storage systems and NoSQL databases. It processes the contents into vector embeddings and subsequently ingests these embeddings into vector databases for efficient similarity searches. Neum AI offers a comprehensive RAG solution that easily integrates with an application, reducing the time and effort spent on data connectors, embedding models, and vector databases.
Exciting Features of Neum AI
- High Throughput Distributed Architecture: This feature enables handling billions of data points effectively. It supports high parallelization levels to optimize embedding generation and ingestion, thereby increasing efficiency.
- Built-in Data Connectors: Neum AI comes with pre-built connectors to widely-used data sources, embedding services, and vector stores, streamlining the setup process.
- Real-time Data Synchronization: Ensures that your data is always current and updated, leading to reliable and consistent outputs.
- Customizable Data Pre-processing: Offers flexibility in loading, chunking, and selecting data according to your specific needs.
- Cohesive Data Management: Provides seamless hybrid retrieval with metadata, as the platform automatically enriches and tracks metadata to enhance the retrieval experience.
Communication Channels
The Neum AI team welcomes interactions and can be contacted via email ([email protected]), on Discord, or through a scheduled call.
Getting Started with Neum AI
Neum AI Cloud
Prospective users can sign up at dashboard.neum.ai. The Neum AI Cloud, which supports large-scale and distributed architecture, is equipped to handle millions of documents transformed through vector embedding. Detailed features can be reviewed under Cloud vs Local.
Local Development
Getting started locally is straightforward with the neumai package, which you can install by executing:
pip install neumai
Upon installation, you can create your first data pipelines by following a quickstart guide. Essentially, a pipeline consists of sources for data extraction, an embed connector for vectorization, and a sink connector for vector storage. Code snippets provided in the documentation demonstrate creating and running different pipelines, such as those involving a website or Postgres connector, and highlight how to publish these projects to Neum Cloud.
Self-hosting
For deployment on your own cloud setup, the Neum AI team provides consultation via email. There is also a sample backend architecture available on GitHub for reference.
Available Connectors
Neum AI supports a variety of connectors. For sourcing data, connectors include Postgres, hosted files, websites, S3, Azure Blob, Sharepoint, Singlestore, and Supabase Storage. Embed Connectors include OpenAI embeddings and Azure OpenAI embeddings, while Sink Connectors range from Supabase Postgres, Weaviate, to Qdrant, Pinecone, and Singlestore. An evolving doc page keeps this list current.
Neum AI Roadmap
The roadmap for Neum AI focuses on expanding both connectors and search capabilities. Planned future developments include integrating new data sources like MySQL and Google Drive, offering advanced features like smart routing, and enhancing query retrieval through metadata attributes. Experimental features such as async metadata augmentation and structured search connectors are under consideration.
Neum Tools
Additional resources and tools related to Neum AI, like neumai-tools, are available. These tools assist in pre-processing tasks such as loading and chunking data prior to the generation of vector embeddings.
Neum AI's robust feature set and comprehensive tools make it an invaluable platform for developers seeking to harness the power of their data.