sap-rap-ai-accelerator

SAP RAP AI Evaluation and Governance Accelerator is a repository of reusable SAP ABAP components and integration patterns for adding evaluated, governed AI capabilities to RAP business objects. It separates generic model connectivity from business-specific validation and prompt logic. RAP applications can reuse the same AI client for explanations, summaries, recommendations, classification, enrichment, and draft-aware review actions without rebuilding HTTP and authentication handling for every business object. The included recursive product hierarchy is a working sample, not a required data model or dependency.

The architecture keeps RAP responsible for authorization, transactional consistency, validation, drafts, persistence, and approval. AI output is advisory until application-specific ABAP code validates and a user accepts it. ZCL_RAP_AI_GOVERNANCE evaluates structured AI recommendations against ABAP-authorized target IDs, allow-listed actions, a fixed recommendation contract, human-review policy, and deterministic acceptance thresholds. It returns grounding, contract, safety, and overall scores with stable rejection codes. The class never persists or approves a recommendation.

ZCL_RAP_AI_CLIENT provides OpenAI-compatible chat-completions requests, SAP HTTP destination or direct HTTPS connectivity, Azure OpenAI deployment URL substitution, api-key and bearer-token header support, JSON escaping and response extraction, and rejection of non-HTTPS direct URLs and embedded URL credentials. This project is a learning accelerator.