behzodhalil

Maybe the idea doesn't work. Maybe no one has executed it well enough yet.

Crowded means unsolved, not worthless

you're right that it's never accurate. neither are we though - people under-report their own intake by 20–40% in actual studies. the photo just has to beat "didn't log anything", and it does.

Fair prior. That's exactly why every number is editable - I treat the model as a quick draft, not a source of truth.

Fair - I hate touching the phone between sets too. Right now the app keeps it minimal: photo for meals, quick-tap for workouts. But "zero effort while exercising" is the direction. What would work best for you - logging the whole workout once at the end, or voice/watch input mid-session? Asking because this decides what I build next.

Ha, that one's on me, not the AI - I made it in Canva myself. Apparently I vibecode by hand too.

Yeah, this is exactly the part I distrusted in other apps too.

So I didn't try to fake it. The model guesses the portion, then you adjust the grams with a slider and the macros rescale live - same "you can always correct it" idea from the post.

Oil and dressing the camera just can't see. That's the honest gap: right now it doesn't add them, so if you know it was cooked in oil, you nudge it up. Rather be approximate and honest than confidently wrong.

Where do the numbers feel most off for you? That's the edge case I want.

No OCR actually - just a vision model doing the recognition, one prompt, JSON out (Qwen3.6-27B on Groq).

I didn't chase some clever accuracy trick, because like you said, a photo alone can only get you so far. The design bet was different: make it fast and make every number editable, so the friction stays low and you fix the guess instead of trusting it.

It nails simple plates and gets confused by messy/mixed ones — that's the part I'm still working on.

Hi, HN I kept giving up on calorie tracking because trying in every meal was too much work. So I tried building something where I just take a photo and it estimates what’s on the plate. It’s honestly not perfect - it does well with simple meals and gets confused by messy or mixed plates, so you can always correct it. But for me it lowered the friction enough that I actually kept logging.

Under the hood it’s a vision model for the food recognition, wrapped in Kotlin Multiplatform (iOS and Android) with a small Ktor backed.

I’d love feedback - especially where the recognition fails for you, so I can see the real edge cases. Thanks for checking it out.