Proof point → Financial transparency starts with being able to read the data banks actually export - currency symbols, broken columns and all.
🩻 Problem
Bank statement exports are hostile data: currency symbols inside numbers, inconsistent columns, unlabeled transfers. Anyone - a person, a startup, or a fund - that wants to understand its cash flows has to clean that mess first.
🔨 Solution
A focused Python CLI for Kuda (Nigerian neobank) spend-account statements:
Architecture Overview
- Cleaning pipeline - pandas with currency-symbol stripping, fuzzy column matching, and category normalization through a maintained
CATEGORY_MAP, plus balance-integrity checks. - Analytics - statistics, matplotlib time series, category and counterparty breakdowns.
- Anomaly detection - 3-sigma flagging of unusual outflows.
- Forecasting - scikit-learn linear regression projecting 30-day net cash flow.
- An argparse CLI documented feature-by-feature in a README more thorough than the tool is large - including a caveat about Kuda’s own export quirks.
📜 Philosophy
Scratch your own itch, but document like a public product: the README anticipates the next user’s setup problems. Small tools earn trust through honesty about their data sources.
🎓 Key learnings
- Real-world financial data wrangling - the gap between “CSV” and usable is where most analytics projects die.
- Lightweight anomaly detection and forecasting that a non-data-scientist can run and read.
📈 Output & impact
- A working open tool (MIT) for a Nigerian neobank’s export format - infrastructure for financial literacy at the individual level.
🌍 Why this matters
Access to Capital. Lending, grant monitoring, and financial compliance all run on one skill: parsing, validating, and analyzing messy Nigerian financial data until it tells the truth. Funding dashboards and credit-information systems are this tool, scaled up.
🚀 Hire me
Drowning in financial data that won’t behave? Let’s talk → · See also: ISP Financial Model · The thesis