Banker’s Book

A personal finance controller with ML-based reconciliation.

Overview

Connects financial records, reconciles activity, flags unusual transactions, classifies exceptions and projects the expected cash position. Matching is hybrid: exact rules for the clear cases, a trained model for the uncertain ones, and a rule-based fallback.

All data shown is synthetic; the scores are on the project’s own benchmark data.

How it works

CatBoost match model with an F1 score of 97.28%, and an 11-class exception model with a macro F1 of 88.98%.
Anomaly detection with rules and Isolation Forest; 7-day and 30-day cash forecasts.
FastAPI backend with a React and TypeScript dashboard.

Built with

Python
FastAPI
CatBoost
React
TypeScript
SQLite