GroundX: A Powerful and Secure Platform for Building Trustworthy RAG Applications

GroundX is an end-to-end retrieval engine that enables developers to build trustworthy Retrieval Augmented Generation (RAG) applications. GroundX allows users to leverage its power in secure environments with no external dependencies. The platform is trusted by organizations like Air France, Dartmouth, and Samsung and has ingested over 2 billion tokens.

Here's a comprehensive discussion of GroundX:

What is GroundX?

GroundX is a feature-complete retrieval engine offered as a service. It offers various tools and services needed to build robust retrieval engines. GroundX provides out-of-the-box or tailored retrieval solutions depending on customer needs and document complexity and can be readily applied to RAG. GroundX was developed alongside LLM providers like OpenAI, Google, and Meta. The platform is designed to tackle one of the most difficult challenges of working with LLMs: hallucinations.

GroundX combines three interlocking systems:

• GroundX Ingest: This system uses a state-of-the-art vision model trained on over 1 million pages of enterprise documents to understand documents and convert them into a queryable representation designed to be understood by LLMs. This system is built to handle visually complex documents in various formats.
• GroundX Store: This component offers secure, encrypted storage for source files, semantic objects, and vectors, ensuring data protection.
• GroundX Search: This system combines text and vector search with a fine-tuned re-ranker model built on OpenSearch to deliver precise, enterprise-grade results. This custom configuration of OpenSearch allows documents processed by the ingest service to be queried against via natural language queries.

Combating Hallucinations with GroundX

LLMs tend to "hallucinate" or generate fabricated information when they don’t know the right answer. GroundX handles hallucinations by ensuring the LLM is provided with high-quality, structured data. GroundX utilizes a proprietary ingestion pipeline which consists of a fine-tuned computer vision model and a pipeline of LLMs to convert documents into clean, simple LLM-ready data. The GroundX ingestion pipeline does the following:

• Converts and cleans documents into an LLM-friendly format
• Dynamically chunks content at natural breakpoints to preserve the context
• Creates contextual metadata for each chunk, providing the LLM with additional information

GroundX On-Premise

GroundX On-Prem is a Kubernetes-deployable instance of GroundX that allows users to leverage GroundX within hardened and secure, even air-gapped, environments. It consists of two key steps:

1. (Optional) Creation of infrastructure on AWS via Terraform
2. Deployment of GroundX onto Kubernetes via Helm

GroundX On-Prem is currently in open beta.

Benefits of GroundX:

• Speed: GroundX APIs are designed to save developers time, allowing them to build RAG apps up to 10 times faster than other tools.
• Accuracy: GroundX greatly outperformed LangChain/Pinecone and LlamaIndex in a head-to-head accuracy test, achieving 98% accuracy across 1,000+ pages of complex documents.
• Security: GroundX prioritizes enterprise security with bank-level encryption. Data is encrypted at rest and during transmission to an LLM, and each corpus is stored in virtual isolation.

GroundX is a powerful platform that allows developers to build trustworthy RAG applications. By focusing on data quality, search accuracy, and security, GroundX allows organizations to effectively and securely harness the power of LLMs.

Link: https://www.eyelevel.ai/post/most-accurate-rag

Documentation: https://docs.eyelevel.ai/documentation/fundamentals/welcome

Github: https://github.com/eyelevelai/groundx-on-prem