Goal
- Learn how to give Claude prompts in a different environment on my project
- Figure out how to force AI into treating this development as a mobile app (this was not the case in the last project)
- Expect a data overlay outcome on my map
- Map out the workflow for this project for the rest of the summer (create a timeline)
Workflow
The Wet Dog Weather team meets with the interns once a week to discuss the projects we’re working on. We all have different projects focused on specific areas of weather knowledge, but we are essentially using the same resources to develop our visions. Given that we use the same datasets and resources, we often encounter similar roadblocks along the way. For this project, one of the interns has been experiencing some difficulty getting Terrier overlays to display. This was intriguing to me because I wrestled with this when I developed my smoke and AQI project. Considering I haven’t focused on integrating Terrier into my project at this point, I decided to listen in to see if I could avoid this happening on my end. Additionally, I was guided on how to export my project from Claude Design to VS Code and how to develop with Claude in that environment.
As I imported TerrierWeb into VSCode, the same errors happened on my end as they did for the other intern. When I tried to run Terrier, I asked Claude to plot temperature data on MapLibre without integrating anything related to my project yet–to see if it worked. Unfortunately, it didn’t, but AI offered the solution to stop Vite from prebundling Terrier and breaking it. At this moment, everything is working properly.
I’m working on shifting my project from a web style to a mobile app style. This involves me asking AI how to capture the window in the correct dimensions while making any sliders and buttons proportional. This has been a challenge, but I expect to have this ready by the next meeting.
Prompts Used
A lot of my interaction with AI has involved asking it what to run and fixing anything that goes wrong along the way. I installed Claude as an assistant in VS Code and set the preference so it doesn’t push anything until I give it permission. Although I have been working with Claude for the past month, I haven’t used it this way before, so I am excited to see what it can do in VS Code. I am exercising caution with it, as when I tried the same approach with Microsoft Copilot, it automatically wrote and deleted code without my permission when I was working on the smoke and AQI project.
For this project, I give Claude prompts like “I want temperature data to plot on the MapLibre map, can you do this?” and the AI runs the commands necessary to make this happen. A lot of next week will be spent experimenting with Claude to see what’s possible in the VS Code environment.
What Worked
I was able to get Terrier running and integrate it into my FishCast project, so that is a win! This is the next big step for my project because it will enable me to plot all the variables required for my fishing index on the map. Additionally, it will output my forecast outlooks in real time.
What Didn’t Work
I am having trouble getting Claude to understand my project in mobile format at the moment. This will be another challenge I need to address to obtain the results I want with AI, and the next fix I need to make before a brief presentation.
Lessons Learned
I tried the “design first, logistics next” approach, but I’m having trouble reacclimating to the VS Code environment. This will take practice, and I’m excited to see the results after this week’s progress.