Small Steps, Big Logic: How Dayton-Phoenix Cracked the Scheduling Code Using AI

Written by Jon Reneberg, Lead Digital Manufacturing Specialist, Purdue MEP

If you’ve spent any time on a shop floor, you know the scheduling puzzle. It’s that daily ritual where production managers try to balance a mountain of orders against a limited set of machines, varying operator skill sets, and the inevitable hot job that just arrived.

For many small to medium manufacturers, this puzzle is usually solved with a pencil and paper, or a complex spreadsheet of intertwined formulas and a lot of copy-pasting. You sprinkle in a little intuition, and you’ve got yourself a schedule, then you rinse and repeat the next day. But as I’ve written before, the journey to Industry 4.0 starts with small steps. Sometimes those steps involve leaning into technologies that seem, at first glance, a bit intimidating.

I recently sat down with Tim Somers from Dayton-Phoenix to talk about a project that illustrates this perfectly. Tim and his team did something that would have seemed impossible five years ago: they built a custom automated production scheduling tool written in Python code using AI.

The best part is that they didn’t know how to program when they started.

Vibe Coding - Using AI as a Coding Partner

“Vibe Coding” is what the cool kids are calling it these days. The term refers to using AI as a programming tool, either by asking AI to create code via common AI chatbots like ChatGPT, Copilot, or Gemini; or by integrating with popular coding platforms. When people talk about using AI to write code, they often think it’s a one-prompt solution. While you can ask an AI chatbot to program something for you in one shot, it gets much more complex when you add specific manufacturing parameters. Tim didn't just write a single prompt, he and his intern spent three months iterating, testing, and refining. They had to work with the AI to ensure the code respected their specific shop floor constraints, like which machines handle tops versus bottoms for their locomotive breaking resistors.

Of course, this still wasn’t an easy task. Each round of iteration involved testing the output on the production floor, getting feedback, and trying again.  Repeat this cycle over and over until everything is just right.  Through the course of it, Tim learned that the production schedule had way more variables than they initially understood. Realizing this meant heading back to the refining stage. Once it was in a condition everyone agreed on, they still had the task of validating their new code for use in daily production.

From Manual Counting to Mathematical Certainty

At Dayton-Phoenix, the process was traditionally manual. Every morning, a supervisor would physically count parts in FIFO lanes. Data visibility in the ERP was limited, and production control was essentially highlighting cells pink to mark jobs as released.

Tim’s team knew what they wanted the program to do:

1. Ingest a list of orders.
2. Level-load work across specific machine pairs.
3. Output a graphical Gantt chart and a clear sequence for each operator.

The goal wasn't to let AI make the decisions, but to use AI to build a program that would consistently produce the same optimized result. As Tim noted, a program can be validated and accepted, whereas a chatbot might give you a different answer every day.

The Tinkerer’s Breakthrough: Coding Without a Degree

Instead of cutting a six-figure check for a third-party scheduling module that might not truly understand their unique constraints, Tim and his intern decided to explore. For the engineers reading this, they utilized the Pandas library for advanced data manipulation and Linear Programming concepts to manage multi-variate constraints.

If you’re a CEO or owner who doesn’t know a string from a float, don't let those terms intimidate you. The most important tool Tim used wasn't a specific line of code, it was curiosity and exploration.

The ROI: 50 Hours to Zero

As a leader, it can be tempting to see an employee playing around with AI as a distraction from real work. But the Dayton-Phoenix story proves that time to explore is actually a high-yield R&D investment.

By giving Tim and his intern the time and space to learn, the company gained:

• Massive Overtime Reduction: They went from 50 hours of overtime in the Fab department and 48 hours in Seam Welding down to virtually zero.
• Proactive Visibility: They now have a 10-day look-ahead window to spot material shortages before they become events where everyone scrambles to make things work.
• Cultural Innovation: They proved that innovation isn't something that only happens in Silicon Valley, it happens right here in Indiana.

Why Exploration is the New Competitive Advantage

This is exactly what we mean when we talk about democratizing technology. You don't need a PhD in Computer Science to start using AI to solve real-world problems. You just need a problem worth solving and a culture that encourages a tinkerer’s mindset.

Are You Ready to Prompt the Future?

At Purdue MEP, we focus on helping Indiana manufacturers find these small steps that lead to massive gains. We’ve watched teams like Tim’s transform their operations, and we want to help you do the same.

If you’re ready to move past the hype of AI and start building actual solutions, check out our upcoming training:

• AI: A Game Changer for Spreadsheets: Our flagship course is designed to help leadership teams learn to explore generative AI use by using it to manipulate Excel spreadsheet data to get impressive results. Students learn how to validate responses, create graphs and infographics, and how to explore AI further.
• AI: A Game Changer for Quality Professionals – Level 1 & 2: This interactive course, broken into two levels, introduces quality and manufacturing professionals to practical, responsible AI use in real-world quality systems. Through hands-on exercises and realistic scenarios, participants learn effective prompting, output validation, and risk-aware implementation to gain measurable value from AI without compromising compliance or professional expertise.

Explore Purdue MEP’s workshop page for open AI courses. Interested in Purdue MEP consulting on an AI project for your organization? Reach out to mepsupport@purdue.edu for more information. 

Writer: Jon Reneberg, 317-275-6810, jonreneberg@purdue.edu