TLDR: SmartAPS is an AI-powered conversational system that uses large language models (LLMs) and specialized tools to help operations planners interact with Advanced Planning Systems (APS) using natural language. It aims to reduce the reliance on expensive OR consultants by enabling users to perform complex analyses like “what-if” and “why-not” scenarios, query plans, and receive recommendations through an intuitive chat interface, significantly cutting down analysis time.
In the complex world of operations management, businesses often rely on sophisticated tools and expert consultants to optimize their planning and decision-making. Operations Research (OR) provides powerful analytical methods, but their application typically demands a high level of specialized knowledge and custom software. This often leads to significant ongoing costs, particularly for Advanced Planning Systems (APS), which are crucial for defining, creating, and interpreting operational plans.
A new system, SmartAPS, aims to change this by making these powerful OR tools and APS more accessible and user-friendly. Developed by researchers from Huawei Technologies Canada, the University of British Columbia, and Huawei Cloud Computing Technologies, SmartAPS is a conversational system designed to reduce the dependency of operations planners on OR consultants.
SmartAPS offers an intuitive, natural language chat interface. This means operations planners can simply type their questions or requests, such as querying an operational plan, performing counterfactual reasoning (what-if and why-not analyses), receiving recommendations, or executing scenario analysis. The system then processes these natural language inputs and interacts with the underlying OR algorithms and tools.
How SmartAPS Works
At its core, SmartAPS leverages tool-augmented Large Language Models (LLMs). While LLMs are powerful, they sometimes struggle with complex numerical calculations or domain-specific tasks. By integrating with external tools, LLMs can act as intermediaries, translating user requests into executable commands for specialized algorithms. SmartAPS uses a catalog of generic tools, curated by OR consultants, and can even suggest or generate new tools when faced with an unsupported query, involving a human in the loop for refinement.
The system is built with three main modules:
-
Conversation Manager: This module understands the user’s intent, whether it’s a casual chat or a specific request about operations planning. It also refines the tool’s output into a natural language response that fits the context of the ongoing conversation.
-
Tool Retriever: When a user makes a query, this module uses advanced embedding techniques to compare the query with a database of available tools and selects the most relevant one based on semantic similarity.
-
Tool Manager: Once a tool is selected, this module extracts the necessary input parameters from the user’s query and the conversation history. It then executes the tool, often interacting with an OR engine like Huawei Cloud’s OptVerse AI Solver, and returns the results.
SmartAPS supports various types of queries, categorized into five main types: ‘query plan’ (e.g., checking inventory levels), ‘why-not’ (analyzing scenarios with new requirements), ‘what-if’ (analyzing scenarios with data changes), ‘compare plan’ (comparing two different plans), and ‘display plan’ (generating visualizations like tables or charts).
Also Read:
- Smart Warehousing: Using AI and Data Maps to Find Inefficiencies
- Machine Learning Enhances Operating Room Scheduling with Predictive Confidence
Real-World Impact
A case study conducted with Huawei’s supply chain planners demonstrated the significant benefits of SmartAPS. Planners reported that the system enabled them to query plans more efficiently and quickly identify reasons for customer order production delays. The ability to perform ‘why-not’ and ‘what-if’ analyses, which previously required direct consultation with OR experts and could take 1-2 days, was reduced to just a few hours with SmartAPS. This direct access to analytical capabilities empowers planners and streamlines their workflow.
While SmartAPS represents a significant step forward, the researchers acknowledge areas for future development. The system’s capabilities are currently limited by the availability of pre-defined APIs, highlighting the need for more sophisticated methods to automatically generate complex APIs. Additionally, long optimization solver times and the current single-user focus are areas for future enhancements, potentially through incorporating a task manager for parallel job execution and exploring multi-user approaches.
SmartAPS is a promising example of how tool-augmented LLMs can create intuitive interfaces for complex systems like APS, ultimately reducing operational costs and improving decision-making efficiency in operations management. For more details, you can refer to the research paper here.
