Nomi

The problem

Personal finance apps are very good at telling you what you already know. You spent this much on groceries. Your subscriptions went up. None of it tells you what to do next, so the insight dies in the dashboard.

How I approached it

Treat the transaction feed as context rather than content. Classify and enrich transactions, then use LLM reasoning to propose specific actions tied to what actually happened in the account, with the reasoning visible so a suggestion can be judged rather than trusted blindly.

What I built

• Transaction enrichment and classification pipeline
• LLM reasoning layer that turns patterns into proposed actions
• Prototype interface for reviewing and acting on suggestions

Where it stands

Working prototype. It is not a product, and it holds nobody’s real banking credentials.

What it taught me

Nine years of banking apps taught me the hard part is not the model. It is deciding which suggestions are safe to make about someone’s money.