Imagine asking, “Why wasn’t Order A scheduled for production yesterday?” and getting not just an answer, but a causal breakdown, an alternative plan, and a visual comparison — all without involving your operations research (OR) consultant.
That’s exactly what SMARTAPS delivers. Built by Huawei researchers, SMARTAPS is a tool-augmented LLM interface for interacting with Advanced Planning Systems (APS) using natural language. It doesn’t try to replace optimization solvers — it simply makes them accessible. In doing so, it redefines how planners interact with complex decision-making models.
From Bottlenecks to Chatbots
Today’s APS software (like SAP S/4HANA or Oracle Fusion) is powerful, but requires expert configuration and costly maintenance. Analysts often wait days for OR consultants to run what-if analyses or test plan revisions. SMARTAPS aims to remove that bottleneck.
Instead of requiring a user to navigate GUIs or understand the intricacies of optimization logic, SMARTAPS provides a chat interface — powered by MISTRAL-7B — that lets planners ask questions, test hypotheticals, and explore alternatives using simple language.
What makes it work? Not magic. Three modular components:
Component | Role |
---|---|
Conversation Manager | Detects user intent (casual vs operational), refines tool outputs using context |
Tool Retriever | Finds the right API to call using embedding-based similarity search (via BGE + ChromaDB) |
Tool Manager | Parses parameters, fills missing info from history, and executes the right tool |
Tools, Not Tricks
Every supported operation is defined in advance as a tool contract. These contracts aren’t code alone — they include:
- A description + example queries (for retrieval)
- Input/output schemas (for execution)
- Template outputs (for response refinement)
For instance, a “Why-not” tool might answer “Why can’t we use the Vancouver plant today?”, while a “What-if” tool could simulate what happens if 100kg of rubber arrives early.
If the planner’s query maps to a known tool, SMARTAPS extracts parameters and executes it. If not, the system flags the gap — prompting tool creation by an OR expert. It’s a clever hybrid: AI for accessibility, humans for domain adaptation.
Real-World Results: Less Waiting, More Planning
In a Huawei production planning pilot:
- Planners previously needed 1–2 days to get feedback from consultants.
- With SMARTAPS, scenario analysis dropped to a few hours.
- Top use cases: identifying delay causes, testing alternative plans, and feasibility relaxation.
Critically, planners didn’t just query plans. They explored them. SMARTAPS turned passive reports into conversational simulations, letting non-technical users interrogate optimization logic on their own terms.
Where SMARTAPS Falls Short
SMARTAPS is powerful — but not omnipotent. It’s bound by its tool catalog. If there’s no existing tool (API), the system can’t help — yet.
Also, optimization solvers still take time. In large operations, jobs often run overnight. The SMARTAPS team notes that future versions will include task managers to track these async jobs.
Lastly, it’s a single-user interface. In reality, supply chain teams have multiple planners with conflicting objectives. Supporting collaborative multi-agent reasoning will be essential.
A Blueprint for Tool-Augmented Automation
SMARTAPS isn’t just a chatbot for supply chains. It’s a reference architecture for business automation:
- Modular: Every operation is a self-contained API.
- Retrieval-aware: No brittle if-else chains — semantic search drives control flow.
- Contextual: Tool outputs are refined by LLMs based on conversation history.
This paradigm — tools + LLM glue — has clear implications across finance, logistics, HR, and beyond. For companies like Cognaptus, it provides a practical path:
Let domain experts define the logic; let AI make it usable.
Cognaptus: Automate the Present, Incubate the Future