# The loan book

> Salary-deduction lending across three parties: the agent who originates, the lender who funds, and the co-operative that collects.

**Sector:** Salary-deduction lender
**Year:** 2026
**Role:** Team of three
**Stack:** Next.js, FastAPI, PostgreSQL, Celery, Redis
**Runs on:** AWS
**What it does:** Reconciliation built in — Angkasa and the bank

## The problem

Salary-deduction lending runs across three parties: the agent who originates, the lender who funds, and the co-operative that collects through payroll. Coordinating that on spreadsheets means reconciling by hand against both the bank and the Angkasa deduction file, and until that reconciliation is done nobody knows the true position of the book.

## What we built

One system covering the whole arrangement. Agents originate, the lender approves and funds, and the co-operative side is served by a direct integration with the Angkasa system for deduction files. Reconciliation against both Angkasa and the bank runs inside the platform rather than in spreadsheets afterwards, with document handling and background processing built in.

## Outcome

- Reconciliation against Angkasa and the bank cut from two weeks of work by two people to a routine inside the system.
- More than 500 loans under management, live since mid-2026.
- Agents, lender and co-operative working in one system instead of exchanging spreadsheets.

## The long version

Salary-deduction lending has an unusual shape: the borrower never really pays
the lender, the employer does, through a co-operative that runs the deduction.
That makes reconciliation the hard part of the business rather than an
accounting afterthought, because money arrives in bulk from a third party and
has to be matched back to individual loans.

Doing that on spreadsheets took two people two weeks. It is exactly the kind of
work that looks like admin and is actually the control system for the whole
book: until it is finished, nobody knows which loans are current. Pulling the
Angkasa deduction data in directly and reconciling it against the bank inside
the platform is the single change that matters here.

The rest of the system exists to keep that data honest at the point it is
created, so agents originate into the same platform the lender approves in and
there is one record of a loan rather than three versions of it.


[Overcode](https://overcode.io/) — AI workflows and custom systems. This study as HTML: https://overcode.io/work/loan-management-system/