AI Smart Expense Tracker

A Flutter client and FastAPI backend built around an “AI hook” — start, end and error callbacks modelled on LangChain’s callback lifecycle — that drives loading, structured JSON and error states.

A small project for practising how an app should behave around an LLM call: show progress, accept only structured JSON, and fail gracefully.

The backend currently simulates the model’s response; connecting a real LLM is the next step.

What it includes

• Natural-language entryOne sentence can produce several expenses.
• AI hook lifecycleStart / end / error callbacks drive the UI states.
• Strict JSON contractItem, category, amount, currency.
• Graceful failureAn error state with an offline fallback view.
• Real LLM parsingPlannedThe backend returns a simulated response today.