After adding shared accounts and multi-user roles, I’ve been experimenting with a small but useful addition to Finbley: an AI-based spending analyst that answers questions only from a user’s own financial data.

The goal isn’t financial advice, but helping users understand patterns they already have.

What it does

Answer natural-language questions like:

- “What category did I overspend on last moonth?”

- “How did my food spending change over the past 3 weeks?”

- “Which expenses are most consistent month-to-month?”

Generate short, structured summaries of:

- monthly spending behavior

- recurring expenses

- unusual spikes or drops

Works per account (supports shared accounts + roles)

What it explicitly does NOT do

- No financial advice

- No investment recommendations

- No predictions based on external data

- No access to data outside the user’s account

It’s closer to a query + summarization layer than a “financial advisor.”

How it’s implemented

- Transactions are first aggregated deterministically (SQL + Python)

- Only sanitized summaries (not raw rows) are sent to the LLM

The LLM is used strictly for:

- pattern explanation

- summarization

- natural-language formatting

Responses are constrained by a system prompt that limits scope to finance data interpretation

This keeps the system predictable and avoids hallucinated insights.

Why I built this

Many users don’t want charts — they want answers like:

“What’s going on with my spending?”

This feature reduces the mental work required to interpret raw numbers.