ShedGPT

TLDR: Gave ChatGPT measurements for a shed; it generated code to run 1000 iterations to find the most cost-effective and optimal building strategy.

In my garden, I have an old shed from the previous owners that’s seen better days. Structurally, it's sound, but it looks like something out of an off-season haunted house. As I planned out the project, I realized that the biggest risk wasn’t in fixing any single part of the shed but in potentially wasting all the trim boards through inefficient cuts.

There were 48 trim pieces in total, spread across 20 different lengths, adding up to 187 feet of material. Each replacement board was sold in 8-foot lengths at $1.50 per foot, so maximizing trim per board while minimizing waste was in my financial interest.

I could have taken the time to manually figure this out, but that would take forever. So, I turned to my trusty friend, ChatGPT.

My first approach aimed to maximize my laziness while testing ChatGPT’s visual capabilities. I started by giving it a picture of the shed with annotated measurements of the largest spans. I hoped it might do some math based on the picture and figure out the lengths of each board.

However, my drawing was pretty terrible, and the results weren’t promising. Looking back, I think if I had provided better prompts with specific definitions and a more detailed drawing, it might have worked better. But I was hoping for some true AI magic and decided on a more direct approach.

Next, I created an inventory spreadsheet listing each trim piece, its quantity, and length. I fed this data to ChatGPT to get started. Initially, the results weren’t great. ChatGPT wasn’t optimizing the boards and was just finding random boards to fill the leftover pieces. It resulted in 48 trim pieces requiring nearly 40 boards, with most of it being cut-off waste. I knew we could do better.

After defining the problem more clearly and framing it as a “Cutting Stock” problem, we started making progress. ChatGPT generated a Python file based on this problem and my required solution. It ran 1000 simulations of possible combinations and determined the best combinations of boards to minimize waste.

The project required a cumulative total of 187 feet of board, and the optimized solution resulted in 157 inches of total waste, using 25 boards. Here’s the optimized cutting plan:

This project not only saved me significant costs by reducing material waste but also demonstrated the practical power of AI in everyday tasks. By leveraging ChatGPT’s capabilities, I transformed a potentially tedious and error-prone task into a streamlined process. This experience has opened my eyes to the vast possibilities of integrating AI into DIY projects, and I look forward to exploring more innovative applications. If you’re tackling a similar project, consider giving AI a try—it might just surprise you with its efficiency and effectiveness.

Hopefully I’ll get another 20 years out of this shed.

Feel free to share your own experiences or ask questions in the comments. Happy building!