All architectures
Proposed DesignSupply chain / SAPAWSMLOps and CI/CD

Cosmetics Supply Chain AI Copilot Platform (v2)

A proposed target design. Components shown are planned or conceptual unless stated otherwise.

Private, human-centred supply-chain AI for one factory, three warehouses and SAP S/4HANA: an agent hierarchy recommends, people approve, and SAP execution stays authorised and traceable.

Overview

  • SAP S/4HANA as the system of record (MM, PP, SD, QM; optional EWM/TM) behind a controlled integration layer: read-only for a POC, human-approved write-backs for production.
  • Agent hierarchy: a cosmetics industry agent, a company supervisor agent (LangGraph + Flowise + MCP tool gateway), and factory-planning, procurement, warehouse, inventory-optimisation, logistics and quality/compliance agents.
  • FastAPI / MCP tool gateway with schema validation, RBAC/ABAC permissions, an audit trail and human approval checkpoints — no unrestricted SAP writes.
  • Private data and knowledge: S3 data lake, RDS PostgreSQL, vector RAG and an optional knowledge graph.
  • Decision-outcome learning and private MLOps on SageMaker; optional, opt-in industry learning that shares only approved, anonymised derived artefacts.

Business use case

Aims at better forecast accuracy, better stock balance with less waste, lower operating costs and improved supplier reliability — with people in control. These are design goals, not measured results.

My role

Architecture design and documentation, building on the SAP supply-chain copilot designed at FPT Software.