Cent needed a way to take loan applications online, keep applicants' personal data safe, move each application through review, and keep a complete record for compliance. From April to June 2026, as Data Pipeline Architect & Systems Engineer on contract, I designed and built that data system. It is live in production.
The problem
A lender holds some of the most sensitive data there is: identity numbers such as the BVN, employment details, and guarantor information. The system had to keep that data protected at rest, let the right people act at the right stage, stop bad data at the door, and give the compliance team a record they could read without a database login.
What I built
Field-level encryption
Sixteen personal-data fields, including BVN, employment data and guarantor details, are encrypted one by one with AES-256 before they are stored. They are decrypted just in time, only when an authorised person needs to see them, so a leaked table or backup shows nothing readable.
A 5-stage loan workflow
Every application moves through five stages as a state machine: it can only move along allowed transitions, each actor (applicant, guarantor, admin) can only take the actions their role permits, and every change is written to an audit log. An application cannot skip a stage or be approved by the wrong person.
Dual sync: database and compliance archive
The working data lives in Supabase (PostgreSQL). A second, append-only copy is written to Google Sheets as a compliance archive: a 70-column record with links to the supporting documents, readable by reviewers without database access. The two are kept consistent by atomic sync steps, so the archive always matches the database.
Validation at the door
Six validation schemas check every input before it enters the pipeline, so invalid data is rejected at the source instead of being cleaned up later. Rate limiting at the infrastructure level protects the forms from abuse and automated fraud attempts.
Why it matters
The result is a lending pipeline where sensitive data is protected by default, every decision is traceable, and the compliance record builds itself. Specific client details are kept private; this case study describes the architecture only.
Stack
Supabase (PostgreSQL), Google Sheets API, AES-256 field encryption, schema validation, distributed rate limiting.
Need this kind of work done?
I take on data engineering, analytics, machine learning and research projects, and I speak and train teams on data and AI.