joaquimcassano

Made the front-end for it. its usable I guess
its not much, but its honest work at least

almost shipping it, just need to improve the UI a bit and ensure everything works (some front-end components are vibe coded and I dont trust AI that much lol)

Added the API route for ADs.
TODO: add the frontend for it

Added the credits feature, that uses the upvotes and downvotes to determine the amount of credits the user has.
TODO: make it useful a.k.a. implement the user-made ads

Added support for user profile pictures (and some small features)

By default, it fetches your profile picture from gravatar (but accepts custom uploaded images). The images are currently stored as base64 directly onto the mongoDB users collection (should I use smth like cloudflare R2? yes. will I? no)

Implemented comments feature, and idea-specific pages (based on ID).
TODO: add reddit-style identation to comment threads

Added settings page

Many improvements on API + responsivity
TODO: implement profile settings page + profile pictures

Connected the back and front ends, so now it’s functional. Btw, small tip: NEVER trust AI on front-end animations, you’ll end up having to rewrite 95% of the code because of how poorly optimized it is lol

TODO: implement ADs

First devlog
So far I’ve been researching, planning how the ADS system will work and prototyping some pages with AI so I can see how they look, and will refactor them later. I also implemented the authentication (this was by hand tho lol)

completely changed the UI. i think it looks cooler now

It is now usable. The API connects with the front-end and extracts the text from the img. Response modes are also available

The API is 100% functional now. Now just need to connect it to the client, fix the 1000000 warnings I received and we can finally start tweaking the front-end and adding the small features

Finally managed to make flash-attn install. Used a pre-built version instead of compiling and building it locally (it would take hours). Now just have to fix some issues w/ huggingface transformers and we should be good to go

I’ve never seen a DX THIS BAD. Every library or version change I make means +5 minutes of build time. I’ma just let Claude handle the errors and go take a shower at this point

Right now, trying to deploy Deepseek-OCR (https://huggingface.co/deepseek-ai/DeepSeek-OCR) on Modal. It’s not my first time using Modal, but it has been a long time since I last used it, so I don’t remember the library’s syntax

Improved the UI on desktop. before, all elements were stacked exactly like on mobile. Now it looks better

Starting to work on the board analyzing logic. Since using a model that can recognize which piece is which is too complicated, I’ll probably take advantage of the fact that all chess games start the same way, and infer the pieces that way (see image)

I’m working on my first project! This is so exciting. I can’t wait to share more updates as I build.