A light-weight demo showing how watsonx Orchestrate can help QA, compliance, and supply-chain teams trace finished goods back to raw materials across SAP mock data, co-packer logs, warehouse movements, and quality inspections.
Traceability Process
The demo focuses on batch traceability for finished goods and raw materials. Users ask natural-language questions, and watsonx Orchestrate selects one of five deployed tools to retrieve traceability, quality, ingredient, or production-listing data.
1. Natural-language request — A user asks for a batch trace, quality status, quality issues, ingredient usage, or a filtered batch list. Example: Which batches contain Taurine?.
2. watsonx Orchestrate tool selection — Orchestrate maps the request to one of five generic tools: trace batch, check quality status, find quality issues, find batches by ingredient, or list recent batches. There is no workflow engine in this version of the demo.
3. Code Engine service execution — The selected FastAPI service queries IBM Db2, combines SAP mock data with data lake records, applies filtering or confidence logic where needed, and returns a human-readable result.
The Result: QA, compliance, supply-chain, and recall users can retrieve batch genealogy, quality decisions, ingredient usage, and production overviews through one agentic interface instead of manually correlating SAP, co-packer, warehouse, and inspection data.
Natural-language traceability, QA, ingredient, or production-list request.
Intent matching and tool invocation.
Code Engine API queries Db2 and applies search, filter, or scoring logic.
Formatted traceability, quality, ingredient, or production insight.
The demo is designed to make the available data and the supported question types understandable before the user starts testing. Users can combine batch IDs, ingredients, co-packers, dates, quality states, and list limits in natural language.
The SAP mock schema contains SAP_BATCH_LINKS. It indicates whether a finished-goods batch has full SAP-side traceability and stores fields such as batch ID, SAP material number, plant, last SAP update, and notes about incomplete coverage.
This reflects the MB56-style starting point: if SAP has the complete genealogy, the lookup is straightforward. In the demo, this is represented as the 30% path.
The Db2 data lake contains finished-goods production records, raw-material inventory movements, co-packer logs, QA inspections, vendor master data, and customer master data.
This is the 70% fallback path: when SAP coverage is incomplete, the tools correlate production windows, raw-material movement timestamps, co-packer evidence, warehouse locations, and QA results.
Traces one finished-goods batch back to raw materials and explains the confidence of each link.
Ask: “Trace batch RB-FG-20260501-A.”
Looks up inspection result, decision code, measured parameters, inspector, and notes for one batch.
Ask: “What is the quality status of RB-FG-20260504-B?”
Finds batches with quality problems and supports filtering by co-packer, date, and outcome.
Ask: “Show quality issues from RAUCH.”
Finds finished-goods batches that contain a selected ingredient or chemical, with optional co-packer filtering.
Ask: “Which RAUCH batches contain Glucuronolactone?”
Lists recent batches and supports filtering by date, co-packer, and material context.
Ask: “Show last 10 batches with materials.”
No. The tools are generic, parameterized APIs. The same trace tool can handle different batch IDs, and the same ingredient-search tool can handle Taurine, Sugar, Caffeine, Glucuronolactone, and other supported materials.
A new tool is only needed when the user expects a new capability outside the current demo scope, not when the user changes the batch ID, material, co-packer, or date range.
Note: This version uses five dedicated watsonx Orchestrate tools and a Db2 data lake approach. It does not use the draw.io workflow engine from the earlier VIG-Smarty demo.
Cloud-Native Architecture
The demo uses a compact IBM Cloud architecture: watsonx Orchestrate provides the agentic interface, five Code Engine services expose tool APIs, and IBM Db2 stores the SAP mock data and operational data lake records.
watsonx Orchestrate — Provides the agentic user experience and maps user intent to the appropriate generic custom tool. The integration is skill-based, not workflow-engine based.
IBM Code Engine — Hosts five lightweight FastAPI services in the Frankfurt region. Each service has its own API endpoint and can be imported into watsonx Orchestrate as a custom skill.
IBM Db2 on IBM Cloud — Holds the demo records across the SAP mock schema and the data lake schema. The data model supports traceability, quality checks, ingredient searches, vendor context, and customer impact analysis.
Raw data approach — The demo uses operational-style tables loaded from CSV into Db2. The services query the tables directly; there is no machine-learning model training or complex transformation pipeline in the demo scope.
Probabilistic traceability logic — Around 30% of batches represent full SAP coverage. The remaining cases demonstrate how the data lake can infer likely material links using timestamps, co-packer logs, warehouse movements, material type and confidence scoring.
Demo application — The live combined app gives users a single place to try the traceability experience and inspect the capabilities exposed by the tools.
Component Interaction: A user asks a question in the demo application. watsonx Orchestrate selects a tool, the selected Code Engine service queries Db2, and the result is returned as a traceability report, quality status summary, quality issue list, ingredient usage result, or filtered batch overview.
Custom Tools
Instead of one large backend, the demo exposes clear, focused APIs that map directly to user tasks in watsonx Orchestrate. The tools are generic and parameterized, so users can change batch IDs, materials, co-packers, and date ranges without requiring a new tool.
Traces a finished-goods batch to likely raw-material batches, checks SAP coverage, and returns confidence scores with a readable summary.
Looks up inspection status for a selected batch, including decision code, pH, caffeine, sugar values, inspector, and quality notes.
Finds quality holds, failed batches, and co-packer-specific quality issues to support root-cause analysis and incident response.
Finds batches containing a selected ingredient or chemical, such as Taurine, Sugar, Caffeine, or Glucuronolactone, with optional co-packer filtering.
Lists recent batches and supports date, co-packer, and material filters for production monitoring and daily batch reviews.
Uses evidence such as delivery timing, co-packer logs, material type, and warehouse location to explain likely links when explicit parent-child data is incomplete.
Technology Stack
The demo is intentionally simple: deployed services, managed data, raw operational-style tables, and watsonx Orchestrate custom tools.
Agentic tool selection and user-facing automation layer
Serverless hosting for the five custom API services
Managed relational database for SAP mock data and data lake records
Storage option for batch files and static assets in the IBM Cloud setup
Backend implementation for traceability and quality endpoints
Portable deployment package for each microservice
Combines co-packer, inventory, QA, vendor, and customer records
Simple tool interfaces for trace, status, issue, ingredient, and batch-list operations
Ready to Experience It
Open the combined Red Bull demo application and test batch tracking, quality status checks, quality issue investigations, ingredient searches, and filtered batch lists backed by the five watsonx Orchestrate tools.