How eyelevelai/groundx-on-prem works
High-Level Components (Top Layer)
a. Client Interface Layer (Load Balancer & API Gateway)
b. Core Services Layer (Processing Pipeline)
c. Infrastructure Services Layer (Data Storage & Message Queue)
Core Processing Components (Middle Layer)
• Document Ingestion Pipeline:
• Upload Service
• Pre-process Service
• Process Service
• Layout Service (with OCR and Inference sub-components)
• Queue Service
• Search & Ranking Pipeline:
• Summary Service (with API and Inference sub-components)
• Ranker Service (with API and Inference sub-components)
• GroundX Service
• Layout Webhook Service
Infrastructure Services (Bottom Layer)
• Redis Cache (for temporary storage)
• MySQL Database (for persistent storage)
• Kafka (message streaming)
• OpenSearch (search engine)
• MinIO (file storage)
• Neo4j (graph database)
Node Groups (Infrastructure Layer)
Compute resource allocation using distinct zones:
• eyelevel-cpu-only
• eyelevel-cpu-memory
• eyelevel-gpu-layout
• eyelevel-gpu-ranker
• eyelevel-gpu-summary
Connections and Data Flow
• Resource allocation (dash-dot lines)
Kubernetes Components
Essential K8s elements:
• Node pools
• Service accounts
• Role bindings
• Deployments
• Services
Additional Elements
• Resource requirements for each node group
• Scaling configurations
• GPU requirements for specific services
Key Considerations:
Infrastructure Details
• AWS EKS cluster boundaries
• VPC configuration
• Availability zones
• Security group boundaries
High-level architecture overview
This will provide a complete picture of the system while maintaining clarity and usefulness for different audiences and purposes.