From Exam Challenge to Custom App in Less Than 10 Hours: An AI-Powered Project (Without Any Coding Experience)

Last week, I was facing a challenge. As part of my outdoor guide training, I had to pass a tough botany exam: correctly identify 20 random plants from a list of 141. Simply looking at internet resources and botanical books felt tedious and ineffective. I needed an interactive tool tailored to my specific list. Since I couldn't find one, I decided to build it myself.

This is the story of how Vigu Växtspel was born.

The 5-Minute Prototype from Claude

With limited coding experience, I turned to the AI assistant Claude, which is renowned for its code generation. I described my game idea in Swedish and in minutes, I had a working prototype. It was proof that the idea was viable.

Working with Scarcity – Modular Design

The initial AI-generated code was a single, long HTML file. I quickly realized that to add features and fix bugs, I would hit the message limits of the free version. I took this as a challenge and wanted to see how far I could go with the free tools. This led me to apply modular design. I broke the project into logical components and had the AI write a prompt that would allow me to work on the different modules without the AI needing the full history of the other modules:

• main.html: The user interface and main structure with the embedded game logic tying everything together
• plant-data.js: A dedicated file for all plant data (names, categories, plant information, hints).
• image-handler.js: A module to handle all logic for loading and displaying images.

This was a game-changer. Suddenly, the project was organized, and adding new features was no longer a headache. To proceed when I hit capacity limits in Claude, I worked on different sub-tasks using other AI engines like Perplexity, Gemini, and ChatGPT.

The Python-Powered Image Quest

A game needs visuals—a lot of them. Manually downloading images for 141 species would have taken hours, if not days. The solution was automation. I wrote a Python script (AI-assisted, of course) that:

• Read my list of Swedish and Latin plant names.
• Used the open API from iNaturalist (a fantastic citizen science platform) to search for high-quality, Creative Commons licensed images of each species.
• Downloaded the top 10 images for every plant (it actually downloaded more than that).
• Created a clean folder structure and renamed the files systematically.

My script became a tireless assistant, completing hours of manual work in less than half an hour. Then, I wrote a second Python script that read my folder structure and an Excel sheet to automatically generate the index.js file. This file became the game's image library, a perfect map connecting each plant name to its corresponding image files.

Improving the Game by Re-Categorising and Adding More Information

I found that the original categories were too broad for efficient learning, so I decided to break them down into smaller groups. I also wanted to add a description of each species, including how it looks in May–June (the time of the exam) and tips for identifying it. In addition, I added an image carousel to allow users to view multiple pictures in the quiz.

While I used Claude to structure the code, the Gemini Pro trial was a good tool for doing longer pieces of work (processing thousands of rows). I asked it to find the information from scientific databases, species by species, and generate the info. This work would have taken me at least a week, but the AI did it in less than 15 minutes.

Excel as Validation

While AI is a great tool to design things from scratch, it still makes mistakes. When I, for example, recategorized the species or searched for information, it sometimes forgot species or even wrote about species not included in the list. Here, I combined programming with a classic tool: Microsoft Excel. I used a spreadsheet to look up and validate the list of all 141 species, their categories, and other data to ensure completeness. When I found a mistake, I simply asked the AI to solve it.

All in all, the project took less than a week to complete, with an actual working time of less than 10 hours. It would have been an even faster job if I had been willing to subscribe and use the more sophisticated paid tools.

More Than Just a Game

VIGU Växtspel became more than just a study tool. It was a project where I learned about AI as a creative partner, the importance of code architecture, working with scarcity, the power of automation with Python, and how to orchestrate different technologies to build a complete solution.

Key Takeaways

• Working with scarcity is a great teacher. Using the free versions takes more time and requires creative workarounds, but it forced me to learn and implement a better, more structured design for the project and better prompts.
• AI tools are a game-changer for building custom products. The biggest takeaway is that by breaking a big problem into smaller pieces and having the curiosity to learn along the way, you can use these tools to build amazing things very quickly.
• AI is a powerful co-pilot, but a human needs to be the pilot. The process worked so well because I used AI for heavy lifting (coding, research) but used my own judgment and simple tools like Excel to validate the results and ensure quality.
• The B2B software landscape is shifting. The ability to rapidly generate hyper-specific custom tools like this one will dramatically change business software. Instead of buying complex, one-size-fits-all platforms, companies can build exactly what they need, when they need it.
• And most importantly, I passed the exam, and the game was well-received by my fellow students! :)

Answer to the question: The plant species shown is Tripleurospermum inodorum

(Swedish: Baldersbrå, English: Scentless Mayweed)