[croisillon]

i looked at the Snapalabra webapp and... i don't even know where to start commenting on it: as someone who started vibecoding, i see how the UX, images, translations, everything feels like when the slop gets horribly wrong ; of course i could read explanations in the about section but it's a 404

why even post that?

[detectivestory]

thanks for having a look... Did you look at the actual app or just the landing page? You ask why would I post it even though it is not perfect. Well, I personally find it useful (thats why I made it). It helps me with bridging the gap between comprehension and expression. And I am curious to see if others also find it useful. I am in contact with professional translators etc but that is probably not something I would be wise to invest too heavily in if the idea itself is not interesting to other people.

[__jonas]

I gave it a try, my impression is this: I’m looking at an AI generated image, and describing it based on AI generated hints, and my description is judged by an AI. The UI is fairly janky on my phone and not very appealing, though I like how it places the hints over the picture.

The fact that it’s all genAI stuff puts me off, but that might not be the same for everyone.

I think crucially, if I wanted to do this kind of challenge, I would have a better experience just using an LLM chat directly, like “Generate me an image, then judge my description of it in <language>, and give me useful language learning hints according to my level of <level>”

[detectivestory]

Thanks for the feedback. The images come from pixabay, pexels, and my own camera, some will be AI generated, most aren't - they are selected for their suitability in learning certain vocabulary. The hints are all manually added. AI is used for translations (although I am in the process of replacing most translations for the non-beta languages with professional translations) and feedback. If you found the web app UI to be "janky" maybe you could try install the native app for your device.

[__jonas]

I see, thanks for explaining the image and hint selection process. I guess I got unlucky then, as today's image is an AI one, and my impression that the hints were generated came from the fact that there is that 'refresh' button that seems to bring up slight variations of the hints, which I assumed came straight from an LLM.

I'll admit, "janky" is not very useful feedback. The first thing that gave me that impression was a couple of seconds of loading time showing just the background color, then the whole page background flashing white every time after clicking/tapping something in the onboarding. I get this both on iOS Safari and macOS Firefox. On iOS, the page also crashed at some point with the 'a problem repeatedly occured on <url>' message.

Looking at it closer, seems like it's a Flutter web app, so I suppose this is to be expected and the experience is probably indeed better on the iOS/android builds, might be that I was judging too quickly.

[croisillon]

ok maybe i've been too harsh, one thing i like is the "sa" flag for afrikaans language, this one really shows attention to the details, i'll try restarting my router to see if it persists

[croisillon]

i went in it with good faith, started the webapp with a couple languages and clicking wherever my mouse landed, quickly enough it appeared to me this was not produced in good faith and i became very critic of it ; at best i would look at it like a proof of concept you send to a team of real dev and designers to get done properly, but in the current state i would never sign in