Case study · AI product

NotSplitwise.

Log now, split later.

A trip-expense notebook. It saves a note, suggests structured fields, and asks the person to review before marking it ready.

I shaped the capture and review workflow, used Claude Code for implementation, and iterated through independent reviews.

AI ExtractionHuman ReviewNext.jsLocal-first PWA
You write

1240 lunch, Asha paid, me + Ben

AI suggests
Amount
1,240
Currency
Trip default
Paid by
Asha
Shared by
Me, Ben
You confirm. Then it is ready.

Synthetic example.

The workflow

Capture → review → mark ready → export

Four steps. The AI only appears in the first, and it never decides what is valid.

  1. 01

    Capture

    Type or dictate a note. It is saved on the device before any AI call.

  2. 02

    Review

    See what the AI suggested next to your original text. Fix what is unclear.

  3. 03

    Mark ready

    Blocked until the amount, currency, payer and people are valid.

  4. 04

    Export

    Copy a plain-text reference to enter by hand elsewhere.

NotSplitwise capture screen for a trip called Goa weekend. The note “1240 lunch, Asha paid, me + Ben” is typed in, with Me, Asha and Ben selected as sharing and Paid by set to “From the note”.
Capture
NotSplitwise review screen for “860 dinner, paid by card, me + Ben”. An amber box says one thing to fix: Who paid? The capture mentions paying but doesn’t clearly say who. The Paid by field is empty and the Mark ready button says 1 to fix.
Review: the AI could not tell who paid
NotSplitwise export sheet listing two expenses, lunch for INR 1,240 paid by Asha and dinner for INR 860 paid by Ben, split equally between Me and Ben, with shared totals by currency and a Copy text button.
Export: a reference for manual entry

Screens are from the local product with synthetic data and a mocked reader, so no model was called.

  • No Splitwise integration, currency conversion, group sync or settlement engine.
  • Not affiliated with Splitwise. The export is a reference to enter by hand.
Product decisions

Three calls about trust

AI suggests. The person confirms. Each decision protects that line.

Capture

Save the note before asking AI

A model failure or no signal should never lose an expense. The note is stored first; manual completion is always available.

Trust

Keep suggestions apart from confirmation

Unclear payers or currency are flagged, not guessed. A field the person corrected by hand beats any later AI result.

Safety

Preserve corrections, validate again at export

Invalid amount text stays a draft instead of becoming a wrong number. Export re-checks each expense and leaves out anything that changed.

One iteration

Recovery must protect the latest intention

A failed save is kept on the device so it can be offered again. The question was what happens to it when a newer correction saves.

  1. STEP 1

    Edit to ₹1,300

    The save fails. The edit is kept on the device as unsaved.

  2. STEP 2

    Edit to ₹1,400

    This save succeeds. It is now the person’s latest intention.

  3. STEP 3

    Older recovery is cleared

    Once ₹1,400 saves, the older ₹1,300 recovery disappears. It can no longer be offered, so it cannot overwrite ₹1,400.

Values are synthetic. The lesson: when something goes wrong with money, recovery should never quietly bring back an older answer.

Evidence and limits

What is checked, and what is not

Capture, validation, export and recovery paths were independently checked with synthetic data and mocked AI responses. Live-model extraction accuracy has not yet been measured.

The redesign and recovery work is pushed to main (f9ac3aa). This page does not claim a particular deployment.

The line I hold

AI suggests. The person confirms.