Balancer

My wife and I were making major money decisions with partial visibility.

We had data across credit cards, bank accounts, savings, pension, investments, and insurance, but no single place to reason about it.

Real household finance data means privacy first.

The current ingestion pipeline is tuned for Hebrew statement formats, so globalization is planned but not completed yet.

I started by using FamilyBiz and Riseup and writing down exactly where they failed for us.

Then I translated those gaps into roadmap milestones, kept one source of truth in the repo, and used handoff notes so each AI session could continue without context loss.

Balancer ingests statements, normalizes transactions, and runs layered categorization using manual overrides, rules, lexicon matching, heuristics, and ML fallback.

The UI is designed to answer monthly questions quickly: where spending changed, which categories drifted, and what needs action.

I keep keyboard access, visible focus states, and readable contrast in light and dark themes because this is an app we use regularly, not a one-time demo.

It replaced the fragmented tracking flow we had before and gave us a reliable monthly view of what is changing.

A public demo is available so visitors can test the product without exposing our real data.

A key AI miss was the first category-editing UX. It took multiple clicks and forced a blocking rule popup.

I replaced it with a one-click popover flow and documented that as a locked decision so future sessions do not revert it.

AI can generate code fast, but product judgment still has to come from me.

The highest leverage habit is documenting decisions as I go, otherwise the same mistakes return in later sessions.