PESAFLOW

Problem Space

A fragmented, manual status quo

Managing money across Tanzania's fragmented ecosystem is a manual nightmare. Users juggle M-Pesa, Tigo Pesa, Airtel Money, Halopesa, Selcom Pesa plus CRDB, NMB and NBC bank accounts — each sending its own SMS format — making it nearly impossible to see total income, spending, and true account balances in one place.

1. 01

   Transaction alerts arrive in wildly inconsistent formats across 8 different carriers and banks

2. 02

   Conventional finance apps rely on cloud sync, exposing sensitive financial data to third parties

3. 03

   Manual expense logging is tedious and routinely abandoned within days

4. 04

   Provider fees, government levies, and VAT make naive balance math drift from reality

5. 05

   Unreliable mobile connectivity renders cloud-dependent trackers useless in rural areas

Process

How the system was built

I treated the SMS inbox as the primary data source and designed around a single constraint: everything must work with the network off. I built a provider-matching pipeline first, then layered a regex parser engine, deduplication, auto-categorization, and balance reconciliation on top of a reactive local store.

1. Decision 01

   Chose Flutter for true cross-platform reach with native background telephony access

2. Decision 02

   Adopted Riverpod for compile-safe, testable reactive state and stream-driven UI updates

3. Decision 03

   Used Drift (SQLite) for type-safe relational queries and reactive local data streams

4. Decision 04

   Locked wallet balances to the carrier's own post-transaction `Salio` to absorb fees and VAT automatically

5. Decision 05

   Ruled out cloud sync entirely — manual encrypted database export is the only backup path

Delivered System

Capabilities shipped

PesaFlow turns the user's SMS inbox into a live, private ledger. Transaction alerts are intercepted in the background, parsed, de-duplicated, categorized, and reconciled against exact carrier balances — all without an internet connection or an account.

Technical Stack

System architecture

A layered, fully local architecture: a presentation layer driven by Riverpod providers, a business-logic layer of parser, transaction, budgeting, and analytics engines, and a data layer backed entirely by Drift (SQLite). An Android background isolate handles SMS ingestion independent of the UI thread.

Framework

Flutter · Dart

State & Data

Riverpod · Drift (SQLite) · fl_chart

Platform

Android Telephony · SMS Retriever API · Background Isolates

Delivery

Cross-platform (Android / iOS) · Offline-only · Manual DB export

Measured Impact

What the numbers say

PesaFlow replaces days of manual spreadsheet reconciliation with a private, always-on ledger that works with the network switched off — a rare combination of automation, accuracy, and total data ownership.

Retrospective

Notes for the next build

1. R-01

   Treating the SMS inbox as a data source demands defensive parsing — carrier formats change without warning

2. R-02

   Deduplication at the ingest layer is essential; a single alert can be delivered more than once

3. R-03

   Reconciling against the carrier's stated ending balance sidesteps an entire class of rounding and tax errors

4. R-04

   Designing for zero-network operation forced a cleaner separation between UI and data layers