jh941213 / NeumAI_RAG

Neum AI is a best-in-class framework to manage the creation and synchronization of vector embeddings at large scale.

Neum AI

Homepage | Documentation | Blog | Discord | Twitter

Neum AI is a data platform that helps developers leverage their data to contextualize Large Language Models through Retrieval Augmented Generation (RAG) This includes extracting data from existing data sources like document storage and NoSQL, processing the contents into vector embeddings and ingesting the vector embeddings into vector databases for similarity search.

It provides you a comprehensive solution for RAG that can scale with your application and reduce the time spent integrating services like data connectors, embedding models and vector databases.

Features

• 🏭 High throughput distributed architecture to handle billions of data points. Allows high degrees of parallelization to optimize embedding generation and ingestion.
• 🧱 Built-in data connectors to common data sources, embedding services and vector stores.
• 🔄 Real-time synchronization of data sources to ensure your data is always up-to-date.
• ♻ Customizable data pre-processing in the form of loading, chunking and selecting.
• 🤝 Cohesive data management to support hybrid retrieval with metadata. Neum AI automatically augments and tracks metadata to provide rich retrieval experience.

Getting Started

Neum AI Cloud

Sign up today at dashboard.neum.ai. See our quickstart to get started.

The Neum AI Cloud supports a large-scale, distributed architecture to run millions of documents through vector embedding. For the full set of features see: Cloud vs Local

Local Development

Install the neumai package:

pip install neumai

To create your first data pipelines visit our quickstart.

At a high level, a pipeline consists of one or multiple sources to pull data from, one embed connector to vectorize the content, and one sink connector to store said vectors. With this snippet of code we will craft all of these and run a pipeline:

Creating and running a pipeline

  from neumai.DataConnectors.WebsiteConnector import WebsiteConnector
  from neumai.Shared.Selector import Selector
  from neumai.Loaders.HTMLLoader import HTMLLoader
  from neumai.Chunkers.RecursiveChunker import RecursiveChunker
  from neumai.Sources.SourceConnector import SourceConnector
  from neumai.EmbedConnectors import OpenAIEmbed
  from neumai.SinkConnectors import WeaviateSink
  from neumai.Pipelines import Pipeline

  website_connector =  WebsiteConnector(
      url = "https://www.neum.ai/post/retrieval-augmented-generation-at-scale",
      selector = Selector(
          to_metadata=['url']
      )
  )
  source = SourceConnector(
    data_connector = website_connector, 
    loader = HTMLLoader(), 
    chunker = RecursiveChunker()
  )

  openai_embed = OpenAIEmbed(
    api_key = "<OPEN AI KEY>",
  )

  weaviate_sink = WeaviateSink(
    url = "your-weaviate-url",
    api_key = "your-api-key",
    class_name = "your-class-name",
  )

  pipeline = Pipeline(
    sources=[source], 
    embed=openai_embed, 
    sink=weaviate_sink
  )
  pipeline.run()

  results = pipeline.search(
    query="What are the challenges with scaling RAG?", 
    number_of_results=3
  )

  for result in results:
    print(result.metadata)

Available Connectors

For an up-to-date list please visit our docs

Source connectors

1. Postgres
2. Hosted Files
3. Websites
4. S3
5. Azure Blob
6. Sharepoint
7. Singlestore
8. Supabase Storage

Embed Connectors

1. OpenAI embeddings
2. Azure OpenAI embeddings

Sink Connectors

1. Supabase postgres
2. Weaviate
3. Qdrant
4. Pinecone
5. Singlestore

Self-host

If you are interested in deploying Neum AI to your own cloud contact us at [e-mail].

We will publish soon an open-source self-host that leverages the framework's architecture to do high throughput data processing.

Roadmap

Connectors

• MySQL - Source
• GitHub - Source
• Google Drive - Source
• Hugging Face - Embedding
• LanceDB - Sink
• Milvus - Sink
• Chroma - Sink

Search

• Retrieval feedback
• Filter support
• Unified Neum AI filters
• Self-Query Retrieval (w/ Metadata attributes generation)

Extensibility

• Langchain / Llama Index Document to Neum Document converter
• Custom chunking and loading

Experimental

• Async metadata augmentation
• Chat history connector
• Structured (SQL and GraphQL) search connector

Additional tooling for Neum AI can be found here:

• neumai-tools: contains pre-processing tools for loading and chunking data before generating vector embeddings.