← Back to Projects
Full-StackIterating

SnapFuel

Photo-first calorie logging with AI estimation and human confirmation.

I built SnapFuel to reduce food logging friction while keeping the user in control. The core idea is: AI suggests, user confirms, dashboard reflects confirmed values.

Date
2026
Signal
Full-Stack
Build stage
MVP shipped, improving reliability and integrations
Stack
Next.js, TypeScript
fullstackhealth-techvision-apiproduct
Product preview

Project notes

Highlights

What I built

  • End-to-end flow from image upload to confirmed meal persistence.
  • Vision output structured with validation and editable review UI.
  • Data model supports manual + AI-assisted entries side by side.

Architecture

How the system works

  • Client upload triggers analysis endpoint for structured meal estimation.
  • User review step edits calories/macros before save.
  • Dashboard queries confirmed records, not raw model output.

Challenges

What made it hard

  • Model confidence needed transparent communication, not fake precision.
  • RLS policies had to be tuned for user-specific meal access.
  • Timezone boundaries affect daily summaries if not handled explicitly.

Lessons

What I learned

  • Human confirmation is the actual product feature.
  • Schema-first planning made API/UI integration much smoother.

Stack / materials

Next.jsTypeScriptSupabaseOpenAI VisionVercel
  • Garmin-style burn integration is currently abstracted behind an interface.
  • Still balancing UX speed with data correctness safeguards.

Media timeline

Build photos, clips, and process visuals. The goal is to show how the project evolved, not just the final screenshot.

Preview
SnapFuel thumbnail