An AI Finance Assistant Inside the MyFinMate App

The situation

The app already tracked income, expenses, budgets, investments, goals and shared group expenses, with reports and charts.

The data was all there, but getting an answer still meant opening the right screen and setting the right filter. The aim was to let users simply ask.

What we built

A separate agent service beside the existing app, with a chat panel inside it. Each message is first classified, then routed to the right handler.

• Questions: the agent uses read-only tools over the user’s records - expenses, income, budget status, goals, investments, spending trends and a monthly overview.
• Actions: adding or changing a record is drafted as a proposal and shown as an approval card. It runs only after the user approves.
• Multi-step requests: handled as a workflow rather than a single reply.
• Unclear requests: the agent asks a clarifying question instead of guessing.
• Memory: conversations are saved, so a follow-up question keeps its context.

Looking ahead, not just back

A background monitor scans each user’s data and raises a notification for budget overruns, goals at risk, unusual spending, upcoming recurring payments and spending trends.

Guardrails

The agent only ever sees the signed-in user’s data, using the same session as the main app. Rate limits, cost controls and caching keep usage predictable, and an evaluation suite tests the agent’s answers against known cases.