Preparing for the Cursor Refactor

Mar 22, 2026 - Eric Boehlke

The resistor reader was a project that I thought up in college. I wanted to create a camera that could take a picture of a resistor and display the resistance value on a 14 segment display.

In college, I built all of the hardware but I didn’t get around to implementing anything.

A year or so later, I wrote enough code to take pictures and gather test data. I had the whole test suite.

Then the project sat abandoned because I was a bit intimidated and din’t know where to start

In 2025 I got a chatgpt pro subscription and that gave me access to codex. At that time codex could clone a git repo, make changes, and create a PR with its changes. I gave it the idea of what I was trying to do and it helped me create a structure for the project and got up to 50% accuracy on my test suite. HOwever I wasn’t able to get past that point.

Now in 2026 we have been using Cursor at work and I want to see if it can help me get over the last hump of this project. While this is something I could do myself, it seems more fun to learn cursor than to fine tune opencv myself.

So this project has shifted from teaching myself computer vision to figuring out how best to use agentic AI tools.

My current approach is to document what I want everything to look like in ARCHITECTURE.md, how we will test in WORKFLOW.md, and individual steps in STEPS.md. I worked with google gemini to improve these documents until I thought they were as solid as reasonable. I feel like these documents could easily communicate to another human what needs to be done in the project while leaving wiggle room for the implementation to change. I think what will give this approach an advantage over codex is that I documented how to create an interactive visualization application with opencv and how to keep track of test failures over the many expected iterations of fine tuning.

First prompt

“I am building a resistor reading pipeline on a Raspberry Pi Zero. I have defined the architecture, data models, and workflow in the attached markdown files. Please read them carefully. For our first step, let’s execute Phase A from the Steps document: creating models.py with the defined dataclasses and Enums. Please provide the code for models.py.”

linked to all of my documents