DEV Community

Tarun Korat
Tarun Korat

Posted on

I built an AI app that estimates calories from a food photo — here's what I learned

The problem I kept running into

I've tried calorie tracking apps on and off for years, and I always quit within a week. Not because I lacked motivation — because the friction of manual logging kills consistency. Searching a food database, guessing portion sizes, weighing everything on a kitchen scale... by day 4 I'd stop.

So I built Calo — an app where you take a photo of your meal and it estimates calories, protein, carbs, and fat using AI, instead of making you search and log manually.

What it does
📸 Photo-based calorie/macro estimation — snap a photo, get an instant breakdown
🗓️ Personalized 7-day diet plans based on age, weight, height, activity level, and dietary preference (veg/non-veg/vegan)
💬 AI chat coach for meal suggestions and motivation
⏱️ Intermittent fasting timer with milestone alerts
👟 Step tracking via Health Connect
📊 Weekly/monthly analytics
🌍 Multi-language support — just added, still expanding
The hard part: making photo estimation actually useful

This was (and still is) the core challenge. A few honest learnings:

Single, visible foods (a plate of rice, a piece of chicken, a salad) — estimation is reasonably solid.
Mixed dishes — curries, casseroles, anything with hidden oils/sauces — accuracy drops noticeably, because the AI can't see what's mixed in.
Portion size estimation without a reference object (like a hand or utensil in frame) is still the trickiest part to get consistently right.

I'm actively iterating on this — if anyone here has worked on food recognition or portion-estimation models, I'd genuinely love to compare notes.

What's next
Improving mixed-dish accuracy
Expanding language support further based on where downloads are coming from
Refining the AI diet plan generation based on early user feedback
Try it / feedback welcome

It's live on the Play Store — search "Calo" or grab it here: Play Store

Still very early days (small number of downloads so far), so genuinely looking for feedback:

Does an AI-estimated calorie count feel trustworthy enough to use daily, or does it need to be paired with manual verification?
What would make a tracking app like this something you'd actually stick with past week one?

Happy to answer anything about the build — happy to write a deeper technical follow-up on the photo-recognition pipeline if there's interest.

Top comments (0)