Overview
PocketLedger makes personal finance less painful. Import a CSV from any bank, and the app automatically categorises each transaction using an OpenAI-powered classifier. A clean dashboard then surfaces spending trends, monthly budgets, and anomaly alerts.
Architecture
Built with Next.js App Router and a Postgres database managed through Prisma. The AI categorisation pipeline runs as a server action — each new import triggers a batch call to the OpenAI API, which returns structured category labels stored back in the database.
Key Features
- CSV import — supports exports from all major Hungarian and EU banks
- AI categorisation — GPT-4o classifies each transaction on import
- Budget tracking — set monthly limits per category with real-time progress bars
- Anomaly detection — highlights unusual spending compared to your rolling average
Challenges
Handling the wide variety of CSV formats from different banks required a flexible parser that could detect column mappings automatically. A schema-inference step using regex heuristics solved most edge cases.
Results
- Tracks 4+ years of personal transaction history
- Average categorisation accuracy: 94%
- Reduced monthly financial review time from ~2 hours to under 10 minutes