Case study · Backend systems · 2026
A private ledger that turns raw bank text into structured transaction records.
01 / Problem
Bank exports and pasted transaction messages are inconsistent, but financial records need repeatable parsing and strict ownership boundaries.
02 / Decisions
A rule-based parser handles raw text and CSV inputs so the same transaction produces the same structured record.
Application scoping is reinforced with PostgreSQL row-level security, reducing reliance on a single user filter.
Optional insights work from aggregate summaries rather than raw transaction rows, keeping the AI boundary narrow and explicit.
03 / Flow
04 / Reflection
Ledgerly sharpened a principle I now reuse: deterministic systems should own the facts, while AI should operate on a deliberately limited surface.
Back to selected work