# itlogs.food

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Pedro Delfino](https://cerebralvalley.ai/u/pedrodelfino)
- **GitHub:** https://github.com/pdelfino/itlogs.food
- **Demo video:** https://www.youtube.com/watch?v=JwRrb7FgCuk
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3559

Site: https://www.itlogs.food

Logging food means typing every meal, and the typing kills the habit. Photo apps skip the typing but guess the portion: four popular ones logged calories and fat about one-third too low across about 100 meals.

itlogs.food is a hands-free food logger. Wear Ray-Ban Meta glasses and put the food on any digital kitchen scale. The glasses see the food and the scale's display. Muse Spark names the food and reads the grams off the display, so the weight is measured, not estimated, and nothing has to pair with the scale. One row appears per food, with its calories and macros, and one tap sends the meal to Apple Health. It logs what was weighed and recommends nothing.

Reading one photo is the easy part. The work is in deciding what a stream of reads means:
- A frame only costs a model call once the scene is still, and the sharpest recent frame is the one read.
- One confident read logs a food; a doubtful one needs a second that agrees. A new read of the same food corrects its row instead of adding one, and one empty read is treated as a hand over the display.
- In a bowl, each rise on the scale is one ingredient, with only the grams it added, named from what the wearer says or by comparing the bowl before and after. The grams are proven on real recorded reads and one real dish was named from speech; a full bowl of several ingredients has not been run end to end yet.
- A packaged food is identified from its wrapper. Muse Spark copies the brand, flavour and net weight off a live frame, then its web_search tool finds the printed nutrition facts. The numbers are only trusted if the protein matches what the wrapper prints and the calories fit the macros, because a wrong number is worse than a blank.
- A dish with no wrapper and no table row, such as a Brazilian esfiha, gets typical values from the same web search, which also corrected a misheard name: "Brazilian Sijah" became "esfiha de carne". The grams stay exact; those macros are typical values, and the ear says "about".

Where it stands, stated plainly in the README: it works on real glasses in good light, with 27 recorded sessions, 22 of them with reads. It fails in a dim room, where the digits wash out. There is no accuracy figure against ground truth yet, and only one scale has been tried.

You cannot try the glasses path without the glasses, so the repo has a way to check it without them: make replay runs the reads of real recorded sessions through today's logic, with no key and no network, and CI fails if a row changes. AGENTS.md lists a command to check each claim.

Who it is for: people who weigh their food and want the log without the typing, and people who stopped logging because of it.

Built during the event. The first code was written after kickoff in a private repo for an earlier idea and carried into this one on November 1. docs/rules-compliance.org lists the commits and goes through the rules one by one.

## More from Meta Global AI Developer Hackathon

- [gentcod](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3556)
- [Team Lumen](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3557)
- [Prudence](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3558)
- [Altair](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3560)
- [Simulicity IA](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3561)
- [Fab Gym](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3562)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3559. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt