The Semantic Layer as LLM Context
As of 2026-11-07
Reframes SAP Datasphere business-layer modelling (dimensions, facts, the Analytic Model) as literal LLM context, using SAP's own documented five-step JustAsk pipeline: Joule sign-in through IAS, a Destination to the JustAsk service, semantic search over indexed model metadata in a customer-specific Vector DB, LLM-generated OData/SQL, and HANA Gateway execution under principal propagation. Explains why the Analytic Model — SAP's 'go-to analytic consumption entity for both Data Layer & Business Layer' — is the object an LLM-driven query actually reads, and cross-validates the discipline (descriptions as machine input, synonyms, value dictionaries, join cardinality) against Databricks Genie's independent best-practice guidance inside SAP Business Data Cloud. Three exercises rewrite labels for machine legibility, design a scoped analytic model, and diagnose model sprawl. Builds on “AI-Ready Data — Quality, Semantics and Lineage”; feeds “Natural-Language Query over SAP Data — Text-to-SQL Patterns and Limits” and “Embeddings and vector search on HANA Cloud from Analytic Data”.
What you will learn
- Build an Analytic Model scoped to one business question in Data Builder → New Analytic Model and set its business name and measures in the side panel
- Rewrite a technical measure into a description with stated formula, grain, what it does not mean and synonyms
- Trace the five JustAsk pipeline steps and say at which step a wrong-model answer arises
- Triage a wrong or missing JustAsk answer in the fixed order platform prerequisites, model selection, candidate metadata, consolidation
- Declare association cardinality and check that a header-level measure does not double-count after an item join
- Run a 20-question gold set with in-scope, refusal and access cases after every release and sign it off
Module overview
Who this is for. You have built dimensions, facts and analytic models in SAP Datasphere for dashboards and stories. This module reframes the same modelling work for a second consumer that reads it just as literally as SAP Analytics Cloud (SAC): a large language model, through the JustAsk and Joule pipeline. It assumes AI-ready data and feeds natural-language query. By the end you can rewrite model metadata so retrieval picks the right model, triage a "JustAsk gives the wrong answer" ticket in a fixed order, and run a 20-question gold set after every release.
Prerequisites
- AI-Ready Data or equivalent understanding of catalog metadata and lineage
- Hands-on experience building analytic models, dimensions and facts in SAP Datasphere or BW/4HANA
- AI & LLM Fundamentals or working knowledge of context windows and grounding
Outcomes
- Design a business layer object legible to both a human building a story and an LLM generating a query.
- Explain JustAsk's pipeline precisely enough to diagnose why a specific question returns the wrong model's answer.
- Apply cross-platform evidence (SAP JustAsk plus Databricks Genie) to justify semantic-layer investment to a sceptical client.
- Diagnose and remediate analytic-model sprawl as a semantic-layer, not a prompting, problem.
Full module with the Consultant plan — €19.90/month. The full module adds: the decision framework · the end-to-end scenario walkthrough · the KPI scorecard · the anti-patterns · the code blocks · the knowledge check · the diagrams.
Open missions
107 live roles and contracts in SAP AI & Joule right now — see the SAP AI & Joule missions · freelance SAP contracts.