Summary
The world’s most important information doesn’t live on the public Internet and never will. That’s why EyeLevel.ai designed their next generation RAG (retrieval augmented generation) platform to run in the most secure data centers, including air-gapped. In this use case, we’ll explain how to set up a private RAG (“talk” to your docs application) that lives near your secure data in just a few simple steps.
Born from a team at IBM Watson and Deep Research, GroundX makes deploying RAG with enterprise grade security, accuracy and scalability easy.
After installing on OpenShift, GroundX turns advanced RAG into three simple calls: ingest, search and complete with the LLM of your choice. GroundX handles document ingest, parsing, chunking, storage, search and reranking without any extra work for developers.
GroundX is particularly good at understanding complex documents, which cause most RAG systems to fail. It recently topped the DocBench leaderboard for multimodal RAG accuracy, beating OpenAI, Anthropic and even humans.
The performance comes from a unique approach to RAG ingest. GroundX ingest combines a vision model and a VLM (visual language model) fine tuned on nearly 1M pages of enterprise documents across many verticals including health, insurance, finance, supply chain, construction and more.
That allows GroundX to turn a complex document like this:
Into LLM-ready data like this:
GroundX also employs a powerful 3X RAG search built on OpenSearch that combines text, vectors and a micrograph. The platform autoscales to any workload without accuracy loss, efficiently provisions GPUs and CPUs as needed, is designed to work across a wide range of NVIDIA GPUs from L40S to H100 and has industry leading security features such as 256K AES encryption, security by design (minimal attack surface and single data egress) and easily supports your RBAC (role based security) systems.
Key Benefits
The key benefits of of this use-case and chosen components to highlight include:
- Advanced RAG in three simple steps
- Deploy your RAG near your data, in your secure on-premises data center or cloud
- Enterprise grade security and scalability
- Best in class document ingest model
- Connect to the LLM of your choice
- Integrates with many agentic frameworks including as MCP, Crew.ai, SmolAgents, AutoGPT and others
System Architecture
The diagram below is a visual representation of how GroundX can be integrated into new and existing AI applications on OpenShift. First, data from a knowledge base is uploaded to GroundX. Then GroundX can be easily connected to your AI application via a few API calls.
Once connected, GroundX will automatically retrieve information that is relevant to a user's query. This contextual information can be fed to a language model, resulting in more accurate and useful LLM generated output.
GroundX provides a flexible data model that allows developers to create both simple and highly customized applications. You can use recommended aggregate information supplied by GroundX to implement a highly performant RAG system in just a few lines of code, or you can leverage multimodal information supplied by GroundX to implement advanced multimodal applications.
System Components
The following sections define and describe the hardware and software components chosen for this use-case.
Hardware
GroundX is designed to automatically scale via an orchestrated microservice architecture to efficiently leverage both larger and data centers and small servers. GroundX divides resources into five node groups by default, The required resources for each of those node groups, and details about configuring these node groups, can be found here. GroundX employs several tailored models which are specifically designed to parse incoming data and to support the search process. As a result, GroundX requires NVIDIA GPU accelerated compute resources to efficiently serve those models.
Software
Building an an accurate and secure RAG system within a security conscious environment on OpenShift can be accomplished with the following software components:
- Red Hat OpenShift is a comprehensive, enterprise-grade Kubernetes platform that simplifies container orchestration, accelerates application development, and supports modern workloads like AI/ML, DevOps, which can be run on-prem or in the cloud.
The Red Hat OpenShift Component is provided by: Red Hat.
Learn more about Red Hat OpenShift AI on our site.
- Red Hat OpenShift AI is an advanced platform designed to simplify the development, deployment, and scaling of AI/ML applications in containerized and hybrid cloud environments. Built on Red Hat OpenShift, it provides the tools, infrastructure, and integrations necessary to accelerate AI/ML workflows while ensuring scalability, security, and reliability. OpenShift AI handles the model training, tuning, serving and monitoring.
The Red Hat OpenShift AI Component is provided by: Red Hat.
Learn more about Red Hat OpenShift AI on our site.
- GroundX is an end-to-end RAG platform that is designed to handle all major operations required in retrieval augmented generation (RAG) workflows. GroundX handles ingest, parsing, storage, and retrieval, allowing you to upload large amounts of complex documents and retrieve relevant context via natural language queries. This context can be fed to LLM powered applications to provide greater context and promote heightened accuracy.
The GroundX Component is provided by: EyeLevel.
Learn more about GroundX on the Red Hat Catalog GroundX page. Information on installing GroundX onto OpenShift can be found here.
- AI Application component is the interface used by the end user of the product. This might be a chatbot designed to converse with the user, an Agent designed to do some automated set of tasks on the users behalf, or any other manner of LLM powered system. The core paradigm of interfacing with GroundX is to send a natural language query and then receive a textual response describing data that is relevant to that query. This allows for the following design paradigms:
- GroundX context can be retrieved on every user query and Injected throughout the continuum of chat between a user and an LLM, allowing the LLM to leverage this context when answering user queries
- GroundX can be framed as a tool which can be provided to an agent. The agent can then decide to employ GroundX to search a knowledge base and get additional contextual information.
- GroundX can be used to automatically retrieve context based on programmatic inputs, allowing GroundX to aid in automatic reporting or batch processing jobs.
These applications could be implemented using popular frameworks like LangChain and LangGraph, or they can be simply implemented from scratch to serve more bespoke solutions.
Use-Case Workflow
The following describes the workflow for this use-case as shown in the diagram above:
1. Data Gathering
User Role(s): Data Engineer
The data engineer is responsible for gathering data from multiple sources and uploading it to GroundX. The data in this use-case may come from multiple sources, including HTML, JSON, PDFs, CSV, etc. Data gathering is performed outside of the OpenShift cluster.
2. Document Parsing with GroundX
User Role(s): Data Engineer or Application Developer.
GroundX provides three major services, the first one is parsing. GroundX will automatically parse the content of complex PDFs, scans of documents, tables of information, slide shows, JSON data, html data, and a variety of other data types. You can read about GroundX document Ingest here.
The first point of interfacing with GroundX is uploading documents via the API, which is typically performed by Data Engineers or Application Developers.
3. Data Storing with GroundX
User Role(s): None, Automatic process.
GroundX contains microservices running MySQL, Redis, and OpenSearch which automatically store the parsed representation of input documents into a queryable representation which is designed for RAG style search.
4. Context Retrieval with GroundX
User Role(s): Data Scientist, Application Developer.
Once data has been uploaded to GroundX, application developers can interface with that data using simple search functionality. You can read about GroundX search here. We also have guides to how GroundX can be integrated into RAG, agents, and other workflows, which you can read here.
5. OpenShift AI
User Role(s): Data Scientist, MLOps Developer
OpenShift AI will provide the model inference and monitoring. The AI Application will access the LLM served using the endpoint URL provided by OpenShift AI.
6. AI Application Development
User Role(s): Application Developers.
One of the major benefits of GroundX is that it abstracts away much of the complex operations that are required in making performant and scalable RAG based products. As a result, AI enabled applications can be developed by developers who don’t specialize in AI.
