View a markdown version of this page

Prerequisites and limitations - AWS Prescriptive Guidance

Prerequisites and limitations

Prerequisites

AWS Services and Access:

  • Active AWS account with appropriate IAM permissions for deploying and managing Amazon Neptune, Amazon Bedrock, Amazon ECS, AWS Glue Data Catalog, Amazon SageMaker Catalog, Amazon OpenSearch, Amazon S3, Amazon SNS/SQS, and AWS X-Ray

  • Amazon Bedrock model access enabled (e.g., Anthropic Claude Sonnet 4.0) through the Bedrock console

  • Amazon Bedrock AgentCore access (requires Python 3.10+ for development with the starter toolkit)

  • VPC configuration with appropriate subnets, security groups, and network connectivity for ECS services and Neptune clusters

Note: some AWS services aren't available in all AWS Regions. For Region availability, see the Service endpoints and quotas page in the AWS documentation, and choose the link for the service.

Technical Skills and Knowledge:

  • Understanding of W3C Semantic Web standards (RDF, RDFS, OWL, SPARQL, SHACL, R2RML, SWRL)

  • Experience with ontology engineering and knowledge graph design

  • Familiarity with data catalog systems (AWS Glue Data Catalog, Amazon SageMaker Catalog/DataZone)

  • Knowledge of containerization and Amazon ECS deployment patterns

  • Understanding of event-driven architectures (SNS/SQS, Neptune DB Streams)

  • Experience with LLM integration and prompt engineering via Amazon Bedrock

  • Familiarity with observability tools (OpenTelemetry, AWS X-Ray, CloudWatch)

Data and Infrastructure:

  • Existing data sources (relational databases, data warehouses, data lakes, document repositories) with documented schemas

  • Business glossaries, data dictionaries, or domain expertise to inform ontology development

  • Network connectivity between data sources and the semantic layer infrastructure

  • S3 buckets for storing ontology versions, R2RML mappings, and document processing artifacts

Open-Source Tools (Optional but Recommended):

  • Ontop 5.5.0 or later for Virtual Knowledge Graph implementation

  • ELK, HermiT, or other OWL reasoners for inference and consistency checking

  • SHACL validators for constraint enforcement

  • unstructured.io for document processing (or equivalent document parsing capabilities)

Organizational Readiness:

  • Defined personas and responsibilities (Administrators, Data-Focused Practitioners, Business Analysts)

  • Governance processes for ontology approval, versioning, and lifecycle management

  • Human-in-the-loop workflows for semantic validation and domain expert review

Limitations

Amazon Neptune:

  • Cluster volume size limited to 128 TiB per cluster (up to 64 TiB in some regions)

  • Maximum of 15 read replicas per cluster

  • Query throughput scales to 100k+ queries per second but depends on query complexity and instance types

Amazon Bedrock AgentCore:

  • Framework support via starter kit CLI limited to prominent frameworks (Strands, LangChain, CrewAI)

  • Session duration and memory persistence constraints depend on AgentCore service limits

  • Model access and token limits vary by foundation model provider (e.g., Anthropic Claude)

Virtual Knowledge Graph (Ontop):

  • Query performance depends on underlying data source capabilities and R2RML mapping complexity

  • SPARQL-to-SQL translation limited to OWL 2 QL profile for optimal query rewriting

  • Federated queries across heterogeneous sources may experience latency based on network and source system performance

Reasoning Services:

  • OWL reasoning complexity (ELK supports OWL 2 EL profile; HermiT supports OWL 2 DL but with higher computational cost)

  • Reasoning over large A-Box (instance data) may require materialization strategies or incremental reasoning approaches

  • SWRL rule execution performance depends on rule complexity and data volume

Document Processing:

  • File size limits for document ingestion (typically 50MB per file for unstructured.io and similar tools)

  • OCR accuracy for scanned documents depends on image quality and document structure

  • GraphRAG retrieval quality depends on embedding model selection and chunk size optimization

General Constraints:

  • Incremental adoption requires careful planning of ontology scope and R2RML mapping coverage to avoid incomplete virtualization

  • Cross-domain ontology queries may require multiple execution steps, impacting latency

  • Human-in-the-loop workflows introduce approval latency for ontology changes and entity resolution

Product versions

AWS Services:

  • Amazon Neptune Database (1.4.7.0+)

  • Amazon Bedrock (with AgentCore support, requires Python 3.10+)

  • Amazon ECS (Fargate or EC2 launch types)

  • AWS Glue Data Catalog

  • Amazon SageMaker Catalog / Amazon DataZone

  • Amazon OpenSearch Service (3.5+)

  • Amazon S3, SNS, SQS, X-Ray, CloudWatch

Open-Source Components:

  • Ontop: Version 5.5.0 (released February 14, 2026) or later

  • ELK Reasoner: Latest stable version compatible with OWL 2 EL

  • HermiT Reasoner: Latest stable version compatible with OWL 2 DL

  • SHACL Validators: Implementations compatible with W3C SHACL 1.1 or SHACL 1.2 specifications

  • unstructured.io: Latest stable version for document processing

W3C Standards (Specifications):

Note

This architecture is designed to be version-agnostic where possible, leveraging AWS managed services that automatically update and open standards that ensure long-term compatibility. Specific version requirements apply primarily to development tooling (Python 3.10+ for AgentCore) and open-source components (Ontop 5.5.0+).