All projects
PERSONAL BUILD · CONSUMER AI · OCT 2026

Thali Log.

A calorie and protein tracker built for how I actually eat.

01 · Problem

The problem

Most trackers log a banana well and a thali badly. Home-cooked Indian vegetarian meals have no barcode and no standard portion. [Add one sentence on why this matters to me.]

02 · Product

What I built

A phone-first app. You log a meal by voice, photo or text. The model estimates calories and protein for each item, and you can edit the numbers before they count. It installs to the home screen and runs as one page with no build step.

[Add 2 to 3 phone screenshots with sample data.]

03 · Decisions

Decisions I made

One user, one month.

It has a passcode and no accounts. I test it on myself for a month before any growth work.

Keep the model swappable.

Four providers work behind one setting, and token use is logged per day, so I can see what each log costs.

Let the user correct the model.

Every estimate is editable before it is saved. Trust comes from the correction step, not from the first guess.

Know what stops me from selling it.

Before charging anyone it needs real accounts, a database with per-user rows, payments, per-user cost caps, and data-protection compliance for India and the EU.

04 · Judging it

How I will judge it

Do I still log in week three? How often do I correct the estimate? Which meals does it get wrong? [Replace with real numbers when the month ends.]

05 · Status

Status

In test. Results go in the newsletter.