Adapted from the update iNDustry Labs sends the AI Executive Group, a group of manufacturing and logistics leaders across the South Bend–Elkhart region. This was the first of the series.
How can GenAI improve scheduling?
Deterministic scheduling models usually live inside expensive software packages that are hard to adapt. LLMs offer a low-cost route to building a custom scheduling tool in house. They help in four places:
- Defining the objective. Working out which questions the model should actually answer.
- Identifying the approach. Reasoning through which algorithm fits your operational context.
- Generating the script. Producing custom optimization code in the language you prefer.
- Advising on deployment. Thinking through how to pilot the tool and manage the change around it.
This is not theoretical. An engineer at an Indiana manufacturer built a scheduling tool this way and cut lead time by 55%.
How should we communicate with LLMs?
A prompt framework helps. CORE stands for context, output, role and example. Use whichever tool you prefer, but start a fresh chat for the conversation. Here is the framework applied to building the scheduling tool above:
CORE, worked through
- Context. "As a production manager at a 100-employee metal job shop, I schedule daily production, prioritize orders and manage resource constraints. Our process is sequential: cutting on five laser cutters, drilling at five manual stations, welding at six tables. I have data on cycle times, including changeover, and operator productivity."
- Output. "Create a scheduling tool using that operational data and those constraints to maximize profitability."
- Role. "As an expert consultant, ask the essential questions to define the contextual framework. As a data scientist, develop a Python scheduling optimization script. As an internal change management advisor, make sure it can actually be implemented."
- Example. "We want a Gantt chart visualization in addition to the attached Excel schedule output," with your desired schedule uploaded to the chat.
Checking the work
Ask the model to show the exact sources it used, whether external websites or the spreadsheet you attached. Request its chain of thought by adding "show your reasoning step by step," particularly on reasoning tasks. Then ask it to explain why it believes its own output is good, which surfaces assumptions you might otherwise miss.
Getting more than one opinion
Ask for diverse perspectives and criticism by assigning roles, a supervisor's view for instance. Run the answer past a different model and ask it to critique the first one. And use a meta-prompt: give the context, then ask the model to help you write a clearer prompt before you ask the real question.
iNDustry Labs runs the AI Executive Group alongside the AI Leaders of Tomorrow fellowship at the University of Notre Dame.