Added back the API key dialog, so the app uses the user’s key, instead of leaking mine.

Updated the github releases (both for Windows and MacOS), so the app runs straight out the box without needing the user to input their own api key. This was done upon request from the Shipwrights.

Devlog #7: Packaging the app and shipping!

Heyy there! Continuing from what I said on the next devlog, what I did today was finally turn ProjectGen into an actual app that people can download and run on their pc instead of just through VS code, as well as add some new features.

Then, I used PyInstaller to package everything into a proper macOS .app, including the fonts and icon. The biggest problem I ran into was the dock icon not working for some reason. My actual ProjectGen icon would appear for half a second before default ttkbootstrap icon would take its place. After asking AI to help me find a solution, I ended up using PyObjC to keep re-applying my icon while the app is running, which sometimes works and sometimes just doesn’t, but I decided not to give it too much attention. Since, I am on mac, I couldn’t locally create the .exe file using pyinstaller, so I used Github Actions plus a script Chatgpt wrote to create it from there. Then I set up a windows vm in UTM, to check if the .exe file works properly. Thankfully it does, so that’s great! Other than that, there were some errosr with indentation and syntax, which I easily fixed.

Finally, I made a .gitignore, added requirements.txt, uploaded everything to GitHub, and uploaded the packaged .app as a github release so people can download and run ProjectGen without needing Python or VS Code installed. I used AI a lot in this step, because the icon issue was driving me insane and also to create help create the Windows app. That’s it for now, see you on the next devlog!

Devlog #6: Photo upload for the components + copy code button!

Hey there! Following up on what I said in the last devlog, what I did today what add a way to uplaod a photo of your components drawer instead of typing each componetn manually. There’s nwo an “or” and an “upload image” button right next to the components field, so the user can pick whichever way is more convenient for them. I used base64 to then encode the image and send it to the AI along with the rest of the prompt, along with some extra instructions.

When an image is picked, the components field just shows “Image uploaded” so the user can tell the AI will be looking at the photo instead of reading from the entry field, plus I added a red X button next to the upload button so the user can cancel it to pick another image or use txet instead. Other than that, I also added a “Copy code” button to the build guide’s code section, so you don’t have to manually select text out of the text box anymore to copy it.

Later, I found out that command=self._copy_code(code) doesn’t work the way it looks like it would, since the parentheses call the function immediately instead of waiting for a the button to be cliced. To solve that, I wrapped it in a lambda function so it only runs when the button’s actually pressed. I used AI to create the new prompt for the image upload and also to help me understand some errors the terminal was showing. That’s for now, see you on the next devlog!

Devlog #5: Build guide screen, project types, and caching!

Heyyy! I’ve been putting off writing this devlog, since there’s a lot off stuff I need to talk about. So, picking up from where I left off in devlog #4, since I said I’d give the build guide it’s own screen and fix the colors, that’s mostly what this devlog is about.

The build guide now renders as a seperate screen with seperate sections (Overview, Components, Wiring, Build Steps, Code), pulled from the AI call that returns structured json. I also fixed the fact that the build guide screen was using a blue theme instead of my olive green one, which was because I was using bootstyle="primary" that pulls from ttkbootstrap’s default theme instead of my custom styles.

I also added a project type dropdown in the beginning of the home screen, so the user can choose whether they want a hardware or a software project. Picking software, hides the budget field (since software projects are mostly free), and relabels the components field to programming language.

I fixed another annoying bug where clicking “Build this!” twice on the same project, re-called the AI every time, instead of using the old response. To fix tat, I added a small cache keyed my project name, so if it detects that the build button has already been pressed for the specific project, it displays the old instructions instead of creating new ones every time. I also fixed a small bug with the “Back to ideas” button and debugged some typos.

Next time, I’m thinking of letting users upload a photo of their components drawer and having the AI think of ideas using those components. That’s it for now, see you on the next devlog!

Devlog #4: Moving to ttkbootstrap + adding a “Build this!“button

Heyyyy! I was gonna write this devlog yesterday, but it was getting kinda late, so I decided to do it toady.

So here’s what’s new: First of all, I switched the app over from plain tkinter widgets to ttkbootstrap, to basically just make things easier. Before, the “Generate ideas!” button was just a label pretending to be a button, with custom hover colors and click bindings, because MacOS (my operating system) was applying an ugly outdated gray/silverish button with no styling, and I wasn’t able to change it.

The switch was relatively smooth, since for most commands, I just had to change the prefix from tk. to tb. Only problem was that these new widgets don’t take bg=/fg= like my previous ones did, so I had to define named styled in the beginning instead. I also added a difficulty field in the beginning, so the user can control the difficulty level of the project ideas, and also a progress bar along with the cycling messages, so the user actually knows when the Ai is thinking.

I also gave each card a “Build this! button, which once pressed, sends another prompt to the ai asking for a full guide, including components, wiring, build instructions, and code if needed. I ran into some errors along the way, mostly typos, which I debugged with the help of Ai.

Next time I’ll give the build guide it’s own screen instead of sitting underneath the entry fields, and make the colors of the app consistent between all the screens. That’s it for now, see you on the next devlog!

Devlog #3: Displaying each idea in it’s own dedicated card

Hey there! Today, I switched from dumping all he project ideas the AI gave me into one big ScrolledText box, since just a wall of text with no specific layout or format isn’t excactly great UX.

The main thing I changed, was swapping the text box with a scrollable list of widgets (boxes), with each widget having its own colored background, border, a Label for the project name, one for the description and the extra components and lastly one for the time and cost to make.

Since tkinter frames don’t scroll by default, I had to create a seperate canvas with a scrollbar bound to it and the cards embedded inside it. Because cards are real widgets now instead of just plain text you can use .delete(), when the user clicked the button again to regenerate ideas, it also had to clean up the previous cards before drawing the new ones, since otherwise all the ideas would just pile up one after the other.

I ran into quite some bugs along the way, such as small syntax errors and typos, which I used AI to help me debug.

Next time I’ll probably add some buttons for each card, such as a “copy idea” or a “build this one” button. That’s it for now, see you on the next devlog!

Devlog #2: Transferring project to tkinter!

Today I moved the project from a terminal-based script to an actual GUI using the tkinter library, since typing into input() prompts and just reading a wall of terminal text isn’t exactly a great user experience.

The main structure translated pretty easily: text boxes (entry widgets) for components, hours and budget, a “Generate Ideas!” button and an output area for the results.

The most difficult part was keeping the window responsive: calling the AI API directly would freeze the whole window until the response came back, so the request now runs in the background, with the result handed back to the main thread safely using root.after().

I used the same rotating status messages from the terminal version (“Brainstorming…”, “Thinking”, etc.), which now update live in the GUI while it waits.

Because this was my first time using tkinter, I used Claude to get me up to speed with the core concepts (widgets, entries, `.pack(), etc.). Once I wrote the code, I also used it to help me debug it.

Next time I’m probably gonna make the ai output structures json so I can display each project idea in it’s own “card”, and generally work in making the GUI look prettier. That’s it for today, see you on the next devlog!

Devlog #1: Coding MVP

Hey there! First devlog for this project! So the idea of this project is that you write what components you have, how much time you can devote and what your budget is, and AI brainstorms 5 different project ideas you can make that fit those criteria.

The project is coded in Python. First, I generated a new API key from HackAi to use as the brain, and integrated it. Then I wrote 1 master prompt which gets sent to the AI every time and includes the info the user has given (components, time available, budget), along with instructions on what format to return each idea in (name, description, extra components needed, cost, and time estimate).

I used the rich library to make the terminal output look prettier: rendering the AI’s response as formatted markdown instead of plain text, adding colors and even a colored spinner along with some phrases (such as thinking..., brainstorming..., ect.) so the user knows somethings is happening instead of just guessing if the program has frozen or not. I used AI to help make the prompt and debug.

Next up: I’ll probably switch to Tkinter to give it an actual GUI instead of a terminal prompt. Probably gonna have text boxes for components/time/budget, a “Generate Ideas” button, and each idea shown in its own card instead of one long markdown dump.

Follow along to see how it turns out!