Back to feed
Dev.to
Dev.to
7/4/2026
I built a Telegram bot that counts calories from food photos. It confidently called soup "berry compote"

I built a Telegram bot that counts calories from food photos. It confidently called soup "berry compote"

Short summary

Built a Telegram bot that identifies food from photos and calculates calories using vision LLM with structured JSON output. Key insight: contextual reasoning rules (e.g., liquid in bowl = soup) outperform long ingredient lists for accuracy. Shipped with freemium pricing via Telegram Stars.

  • •Vision LLM analyzes food photos to estimate calories, macros, and portion weights on-the-fly
  • •Contextual rules (serving context, texture, liquids) beat training on dish lists for accuracy
  • •Freemium model (5 free analyses + 1 daily) monetized via Telegram Stars; seeking multi-cuisine feedback

Generated with AI, which can make mistakes.

#ai-tools#product-launch
Read full article at Dev.to

Is this a good recommendation for you?

Comments

Failed to load comments. Please try again.

Explore more

I built a BYOK AI calorie tracker because food logging is still too slow
Dev.to
The original title is "Fud AI in one post: open source, BYOK, snap-a-photo logging"
Dev.to
From Pixels to Proteins: Building a Precise Dietary Analysis System with GPT-4o and SAM
Dev.to
I Built an AI App. Eight Months Later, It Became a Skill
Dev.to