Building for Hydro Node

This week's class work was on custom GPTs, so I used it on my own project, Hydro Node. Hydro Node is a dashboard I'm building to keep track of hydroponics and other systems. I used ChatGPT's GPTs to set up an assistant dedicated to the project. Since it's something I'm already working on, I was interested in it, and having a real project to build around made this feel more personal than just trying a tool for an assignment.

The instructions were not perfect at first. I was able to inject some prompts that made the assistant behave differently from what I intended. I went back and fixed the instructions later. That was a useful part of the process because it showed me that getting an assistant set up is not the same as getting it to follow the instructions the way I want.

In Week 3 I didn't get much time to try things outside class, so it was good to get back to building something this week.

I have mentioned before that I am interested in AI research and potentially doing a master's in AI or a related field. Working on the Hydro Node assistant connects with that because it brings AI into a project I'm building myself. An assistant that knows the dashboard could help answer questions about the systems it tracks, instead of me digging through the data by hand.

It also connects to my interest in agent reliability. Seeing the assistant behave differently after those prompts gave me a concrete example to think about. As I continue working on it, I want to pay attention to whether its responses stay useful for the project. Building it is one part, but checking what it gives back is another part I want to spend time on.

My first midterm finished last week, and the next one is in Week 5. I am really locked in for that one and working hard to get a good grade. Midterms are a big priority for me right now, alongside keeping up with my regular coursework.

With the assistant, the challenge was getting the instructions right. I had to revise them after seeing how the prompts affected its behavior. For me, that connects directly to why checking AI output matters instead of assuming the first version is ready.