consciouscalories.vercel.app ↗ LIVE

How I built it

Fast, mobile-first experience: compress on device, analyze with LLM, show structured result.

Phone
  │
  ├─► Take / pick photo (PNG/JPG/WEBP, ≤10 MB)
  │     compress in browser (canvas) — keeps upload fast
  │     preview + detect food vs non-food
  │
  └─► Send to AI
        Gemini 2.5 Pro (via AICredits)
        tolerant JSON parse → 
        {
          dish, calories, protein, carbs, fat, ingredients[]
        }
        │
        ▼
      Show result cards
      dish name · kcal · macros · ingredient list
      (estimate — not medical advice)

No API key in repo — runs as frontend prototype on Vercel with React 19 + Vite.

Why it works for visitors

• No typing — just a photo
• Compression before upload → quick even on slow data
• Non-food detected so you don't get a fake calorie count
• Structured output every time, thanks to tolerant parsing