SCM POC — Architecture Diagram

Platform

Snowflake Data Cloud Platform

Everything — data landing, transformation, semantic modeling, and the SCM Assistant Agent — runs inside a single Snowflake account. Nothing leaves the platform.

Snowflake Data Cloud Platform

01Data pipeline

Raw Data Files

PUT

Internal Stage

COPY INTO

Raw Tables

Snowpark

Curated Tables

Use-case Specific
Semantic Views

02SCM Assistant Agent & consumption

Cortex AI

Supply Chain Orchestrator / Planner Agent

Use-case Specific Sub Agents

Sub-Agent 1

Sub-Agent 2

Sub-Agent 3

Snowflake Cowork

Users

Git Repo for Agent Skills

FETCH

Agent specific skills

End-to-end Data Lineage

Data Quality Monitoring

Sensitive Data Classification

Data Protection Policies

Auto-generated descriptions

AI Guardrails

Role-based Access Controls

Snowflake Horizon Catalog

Legend

How to read this diagram

Stage	What happens here
Raw Data Files → Internal Stage → Raw Tables	Source extract files are staged and loaded via `PUT` / `COPY INTO` into the RAW layer — one table per source extract, every column loaded as text, no transformation. This is the RAW layer in the Data Dictionary.
Snowpark → Curated Tables	Snowpark-based transformation logic builds the typed, cleaned, business-rule-applied, conformed star schema. This spans the RAW_TRUSTED → STAGING → CURATED → AGGREGATE layers in the Data Dictionary.
Use-case Specific Semantic Views	Cortex Analyst semantic views defined on top of the curated star schema — the SEMANTIC layer. These are what the SCM Assistant Agent actually queries; see the Data Dictionary for full column-level detail on all four live views.
Cortex AI — Orchestrator & Sub-Agents	The Supply Chain Orchestrator / Planner Agent (`SCM_AGENT_WITH_RECS`) routes each question to the right use-case-specific sub-agent/tool (purchasing, delivery, demand planning, inventory) — see the Capability Overview page for what each can do.
Git Repo for Agent Skills → FETCH	Agent-specific skills (reusable instructions/behaviors) are version-controlled in a Git repository and fetched by the agent framework, rather than being hardcoded into the agent definition.
Snowflake Cowork → Users	The chat interface end users actually interact with, built on Snowflake Cowork.
Snowflake Horizon Catalog	Platform-level governance running underneath the whole pipeline: end-to-end data lineage, data quality monitoring, sensitive-data classification, data protection policies, auto-generated descriptions, AI guardrails, and role-based access controls (see the Business Rules & Personas page for how RBAC is designed for this pipeline specifically).