Accelerate financial and billing data extraction

Copy link

Extract structured financial and billing data from unstructured documents, such as PDFs or images, using GroundX&reg.

Table of contents

Copy link

• Detailed description
  • See it in action
  • Architecture diagrams
• Requirements
  • Minimum hardware requirements
  • Minimum software requirements
  • Required user permissions
• Deploy
  • Prerequisites
  • Installation
  • Monitor deployment
  • Demo billing extraction
  • Delete
• References
• Technical details
  • GPU configuration for GroundX inference
• Tags

Detailed description

Copy link

For many organizations, critical financial and billing information remains locked inside unstructured formats like scans, PDFs, and images. Extracting this data traditionally requires slow, error-prone manual entry or brittle, template-based OCR systems that break whenever a vendor shifts a column or alters a layout. Processing complex document structures—such as nested tables, multi-page invoices, and diverse document formats—at scale remains a highly complex technical challenge.

This AI quickstart is designed to bypass those hurdles, helping you get up and running quickly with a robust, production-ready extraction pipeline. You will deploy GroundX from EyeLevel to automate billing data extraction within a secure, on-premises AI environment powered by OpenShift AI.

See it in action

Copy link

See a detailed walkthrough of the UI-based quickstart application: Walkthrough

Architecture diagrams

Copy link

Requirements

Copy link

This quickstart was developed and tested on a Red Hat OpenShift® cluster with the following components and resources. These can be considered the minimum requirements.

Minimum hardware requirements

Copy link

Minimum software requirements

Copy link

This quickstart was tested with the following software versions:

Required user permissions

Copy link

The user performing this quickstart should be able to create a project and install both Helm charts. Roles differ by chart:

[!NOTE] A single make -C helm install runs both charts. Use an account that can install billing-operators (typically cluster-admin). If operators are already installed cluster-wide, an admin can install only the workloads chart.

Deploy

Copy link

Deployment uses two Helm umbrella charts, installed in sequence through a Makefile:

By default, GroundX layout and ranker inference run on CPU. Optional GPU settings are documented under Technical details.

Prerequisites

Copy link

The steps assume the following products and tools are already available on the cluster:

1. Red Hat OpenShift Container Platform
2. Red Hat OpenShift Service Mesh
3. Red Hat OpenShift Serverless
4. Red Hat OpenShift AI
5. Authorino (typically installed with OpenShift AI / Service Mesh)
6. Helm 3.x installed locally
7. oc CLI installed and authenticated
8. The eyelevel project/namespace does not already exist

[!NOTE] GPU is optional. Default GroundX inference uses CPU. Install the Node Feature Discovery and NVIDIA GPU operators only if you enable GPU inference (see Technical details).

Installation

Copy link

1. Clone this repo and log in to your cluster:

git clone https://github.com/rh-ai-quickstart/Billing-extraction-with-GroundX.git
cd Billing-extraction-with-GroundX

oc login --token=<user_token> --server=https://api.<openshift_cluster_fqdn>:6443

2. Create the workloads secret file (required — credentials are not stored in git):

cp helm/billing-workloads/secret.example.yaml helm/billing-workloads/secret.yaml

Edit helm/billing-workloads/secret.yaml and set at least these keys under groundx-secret.data:

[!IMPORTANT] GROUNDX_ADMIN_API_KEY can be any UUID you choose — it does not come from GroundX or another provider. Pick any value in UUID format (for example 00000000-0000-0000-0000-000000000001) and use the same value consistently. Do not confuse it with GROUNDX_AGENT_API_KEY, which must be a real OpenAI-compatible API key.

No shell environment variables are required for install. Helm merges secret.yaml into the chart and creates the eyelevel-secret-credentials Kubernetes Secret.

[!NOTE] helm/billing-operators/secret.yaml is NOT OPTIONAL.

make -C helm install only checks that helm/billing-workloads/secret.yaml exists.

3. Review and edit the values files to match your environment:

   • helm/billing-operators/values.yaml — operator toggles, node labels
   • helm/billing-workloads/values.yaml — GroundX config, notebook settings, resource limits

4. Install using the Makefile (recommended):

# From the repo root — installs operators first, then workloads
make -C helm install

Monitor deployment

Copy link

oc get pods -n eyelevel

All pods should reach Running (or Completed for one-shot Jobs).

Demo billing extraction

Copy link

Access the UI

Copy link

1. Open the frontend UI route in the OpenShift console (Networking → Routes), or:

oc get route -n eyelevel -l app.kubernetes.io/component=frontend \
  -o jsonpath='https://{.items[0].spec.host}{"\n"}'

The URL looks like https://billing-workloads-frontend-eyelevel.<cluster_domain>/.

2. Follow the Data Extraction UI walkthrough below to run extraction in the app.

Typical flow: Infrastructure Check → Upload & Process (try AT&T Wireless) → View Extracted Data → Job History.

Delete

Copy link

Remove the deployment using the Makefile:

# From the repo root — uninstalls both charts and deletes the eyelevel project
make -C helm uninstall

This uninstalls the workloads chart first (clearing CRs and finalizers), then the operators chart, then deletes the eyelevel project.

If the project remains, remove it manually:

oc delete project eyelevel

References

Copy link

• GroundX documentation v2.9
• Red Hat OpenShift AI documentation v3.4
• Red Hat OpenShift documentation

Technical details

Copy link

GPU configuration for GroundX inference

Copy link

By default, the billing-workloads chart runs GroundX layout and ranker inference on CPU. There is no separate “suppress GPU” flag — the chart overrides the upstream GroundX defaults (which request a GPU) in helm/billing-workloads/values.yaml:

Layout inference (groundx.layout.inference):

layout:
  inference:
    deviceType: cpu
    resources:
      limits:
        memory: 12Gi
        nvidia.com/gpu: '0'
      requests:
        cpu: 500m
        memory: 2Gi
        nvidia.com/gpu: '0'

Ranker inference (groundx.ranker.inference) — also CPU by default, with fewer workers and more memory than the GPU defaults:

ranker:
  inference:
    deviceType: cpu
    workers: 2
    resources:
      limits:
        memory: 8Gi
        nvidia.com/gpu: '0'
      requests:
        cpu: 1500m
        memory: 4Gi
        nvidia.com/gpu: '0'

To enable GPU inference, set nvidia.com/gpu to '1' and deviceType to cuda (for ranker; layout follows the same pattern). Use a GPU with roughly 24 GB of memory (for example NVIDIA A10, L40S, or A100). See the comments in helm/billing-workloads/values.yaml for the full GPU resource blocks. Nodes labeled for GroundX (gpuLayout / gpuRanker) must have an NVIDIA GPU available, and the NVIDIA GPU operator must be installed.