Goal
- Have an outline for my air quality dashboard
- Go “AI tool shopping”
- Learn how to use/install GitHub
- Figure out my data sources and how I am going to plot them
- Have a workflow plan for my project
Workflow
The first day and a half focused on developing an organizational system for my work and project. I developed a system to log my hours, tasks, notes, and questions throughout the week, making the weekly documentation write-up easier. Then, I reviewed the materials required for my smoke and AQI project and installed the appropriate software accordingly (and for future projects).
To start the project, I needed to install GitHub (desktop version) and VS Code on both my personal laptop and desktop computer. Then, I watched tutorials to learn how to use these tools, as I had never worked with them before. By day two, I had cloned TerrierWeb and started reading and working with the different map examples in the repo in VS Code. I encountered difficulties plotting the maps on a local web server. It didn’t work on my computer, but it did on someone else’s. It’s unclear what happened: a file may have been corrupted, TerrierWeb may not have been properly cloned, or something else may have gone wrong. It still wasn’t working, and figuring out what went wrong has been challenging and time-consuming.
Meanwhile, I have been using a familiar Python environment (I personally like Spyder) to test AWS files and see if I can plot data from them using LLMs (like CoPilot or ChatGPT). According to ChatGPT, I can plot my data in Python and transfer it to TerrierWeb; however, I’m not yet familiar with TerrierWeb, and my computer won’t run the web version in VS Code.
Overall, this week has been a lot of trial and error, as the internship is supposed to be. I have encountered a few barriers along the way, but I have always been able to use my time wisely despite them. I am hopeful that in our weekly meeting, we can discover “what went wrong” and get back on the right track.
Prompts Used
I have spent some time “AI tool shopping” for the summer. I’m hesitant to commit to any particular AI tools until I know what I’ll need for my future projects. However, I found some tools this week that have been helpful for my project development. ChatGPT was the #1 supporting tool for me and my dashboard development. Not only did it give me step-by-step instructions for building this dashboard (based on the prompt), but it also outlined my timeline and suggested the work I needed to do each day to successfully deliver a complete project. Although I didn’t use this specific timeline, I found this feature useful and plan to use it in future projects.
As for my prompts, I started with my explanation that my idea was to build a website for a smoke and AQI dashboard from TerrierWeb. I shared the TerrierWeb repo from GitHub and the Google Doc with the project instructions in the chat. The output was a general summary of how to build my project, with detailed charts showing the steps I need to take and when to take them. Then I kept the conversation going by asking questions about each step. The process worked out well, as I was able to clearly understand what I was doing and what stage of the project I was in.
If I invested in ChatGPT Premium, I could see it being beneficial for long conversations. In the standard version of ChatGPT, there is a technical limit on the length of conversations. This week, it was helpful to have a long conversation about my dashboard project. If I ran into barriers or had questions while following instructions, it was really simple to ask in the same chat rather than explaining myself all over again in a new one. Additionally, I often upload screenshots of what I am working on to see if ChatGPT can direct me to where to go (for example) if I’m not familiar with the software or website I’m working with. This week, I used this feature less, but I can see how it could be helpful in the future, since ChatGPT also has an image upload limit.
Built into VS Code is Microsoft Copilot, which I have experience using for advanced use of Microsoft Word, Excel, and PowerPoint. Personally, I have never been a fan of Copilot. Whenever I had a question about how to use a certain feature in Word, I always had a difficult time figuring out what the AI was telling me to do. The help and support were useless because the directions were incorrect. It seemed like the AI tool was making stuff up, which was very frustrating when I just needed a “quick fix” in the past. Now, in VS Code, it doesn’t seem like the tool is much better, given my experience not being able to display Leaflet or Maplibre examples on the web, even though there are examples in TerrierWeb.
One tool I have read about in the intern-discussions chat (and was also suggested to me) is Claude Code. This week, I’ll use that tool and evaluate whether it’s helpful per the suggestion.
What Worked
As previously mentioned, plotting the data in a Python environment worked well. There are still some uncertainties with this, as I haven’t uploaded to Terrier yet, so I’m not sure how well this will work overall. Python is relatively easy, and I’m used to working with it, so I am curious to see if it will work out.
What Didn’t Work
A lot did not work out as I had hoped. Both Copilot and ChatGPT suggested I had access to something called a Boxer Stack. From what I understand, this is WDW’s cloud-based environment for weather data processing. I’m not sure where I would implement this in VS Code, how it works, or what I would use it for, but the LLMs seem to think there is a use for it. It pretends to get stuck on this point. I asked about this issue, but I sense confusion on the other end of things as well–another conversation for Tuesday.
Also, there was the whole issue with not being able to display the Terrier Web examples in the web version. This is also a point where I feel like I’m “stuck,” as I am not as familiar with GitHub or the VS Code environment, so I am nervous about when to merge my data into Terrier. I am choosing to remain positive about this issue, but it is frustrating when it works on other computers but not mine.
Lessons Learned
Even if I get “stuck,” there is always something else I can be working on. I felt a little discouraged throughout the week because my TerrierWeb examples weren’t plotting. However, I also recognized that this was a multi-step project and that we still had another week before we present our dashboards. It was very helpful that the WDW staff reassured me that this is a trial-and-error internship, and I can follow whatever process works best for me and how AI will (or won’t) cooperate.