Skip to main content

Interoperability & Standards

A true AI Product must be interoperable — able to integrate seamlessly into broader ecosystems of data, software, and other AI Products.
Interoperability is achieved through adherence to standards, protocols, and shared semantics.

Why Interoperability Matters​

  • Composability → AI Products can be combined into pipelines and workflows.
  • Portability → Products can run across different infrastructures and environments.
  • Discoverability → Standards-based metadata enables cross-catalog search.
  • Governance → Standardized formats ease compliance checks and audits.
  • Future-Proofing → Standards protect against obsolescence in rapidly evolving AI ecosystems.

Interoperability Dimensions​

1. Interfaces & APIs​

  • Products must expose well-documented APIs.
  • Prefer open standards (REST, gRPC, GraphQL, OpenAPI).
  • Where possible, align with ML serving protocols (e.g., KFServing, V2 Inference API).

2. Metadata & Semantics​

  • Products must publish metadata in machine-readable standards: JSON-LD, RDF, YAML.
  • Align with BPS, AIPDS, and AIPROD schemas.
  • Support semantic web principles for catalog integration.

3. Data Formats​

  • Inputs and outputs should use standardized encodings (e.g., JSON, Avro, Parquet, ONNX, HuggingFace formats).
  • Multimedia AI Products must support interoperable codecs (PNG, MP4, WAV, etc.).

4. Model Artifacts​

  • Use portable model formats where possible (ONNX, PMML, MLIR).
  • Declare version compatibility with frameworks (PyTorch, TensorFlow, JAX).

5. Governance & Compliance​

  • Support interoperability with compliance standards (ISO/IEC, EU AI Act, NIST).
  • Declare mappings to organizational or industry schemas (e.g., FHIR for healthcare).

Interoperability Levels​

  • Minimal → Proprietary API, custom metadata, limited portability.
  • Standardized → Open API, partial standards adoption.
  • Semantic → Full alignment with interoperable ontologies and catalogs.
  • Composable → Easily combined with other AI Products and Data Products in workflows.

Example​

AI Product: Fraud Detection Classifier

  • Interfaces: REST API with OpenAPI spec.
  • Metadata: JSON-LD, aligned with BPS ontology.
  • Data Formats: Accepts Parquet transactions, outputs JSON risk scores.
  • Model Artifact: ONNX model with declared framework compatibility.
  • Governance: Aligned with ISO/IEC 42001 (AI Management System Standard).
  • Level: Semantic interoperability with downstream risk management products.

Summary​

  • Interoperability is essential for ecosystem integration, trust, and longevity.
  • AI Products must adopt standards for APIs, metadata, data formats, and model artifacts.
  • Higher levels of interoperability enable composability and governance at scale.

Principle: An AI Product that is not interoperable is not sustainable — it becomes an isolated artifact rather than a composable product.