[07]
BANKING TRACKER
█ TIER A · YEAR 2026 · STATUS: LIVE · LANGUAGES: TYPESCRIPT
AI finance autopilot: receipt OCR (Claude vision) + self-learning rules over open banking (PSD2)
[FIG. 1] MISSION
A personal finance control center where AI removes the tedious work. Receipt OCR via Vercel AI Gateway (Claude Haiku 4.5 vision) splits a photographed receipt into line items. Bank transactions (PSD2 / Enable Banking) stream in automatically, and categorization follows a „human → rules → AI” path: each manual correction becomes a rule, the system gradually takes over sign-off and leaves only exceptions to the user. On top of that: envelope budgeting on a „card = category” rule, three money layers (accounts / envelopes / goals) and one „truly available” figure. Next.js 16 + React 19, Drizzle + Neon Postgres, jose+PIN auth, shipped as a PWA. Modules: analytics, debts, goals, subscriptions, finance boards, splits.
[FIG. 2] ARCHITECTURE
hover a block to see its description
[FIG. 2A] RECEIPT → LINE ITEMS
| MILK 2% | 4.99 |
| RYE BREAD | 7.50 |
| BUTTER | 8.99 |
| APPLES 1KG | 5.49 |
RECEIPT AWAITS SCAN
- ŻABKA −23.50░
- ORLEN −180.00░
- NETFLIX −43.00░
→ you sign every transaction by hand — the inbox keeps growing
[FIG. 3] CHALLENGES
[+][CH-01]
PSD2 (Enable Banking) integration — bank OAuth, token exchange, transaction/balance sync, bank sessions.
[+][CH-02]
Self-learning rules as a path to AI — each manual correction becomes a rule, the system gradually takes over categorization and leaves only exceptions to the user.
[+][CH-03]
A „truly available” model — separating real accounts, budget envelopes and goals so one figure tells the truth about liquidity.
[FIG. 4] AI LAYER
AI finance autopilot: receipt OCR (Claude vision) + self-learning rules over open banking (PSD2)
[FIG. 5] GALLERY
SCREENSHOTS INCOMING — see SHOTS-WANTED