← All projectsCase study · Mobile · Android

Biometric Verification at the Point of Service

A telecom operator needed to confirm each customer's identity before processing a transaction. I built the app its agents have used in production since 2022, then rewrote it from scratch for the years ahead.

Client
Telecom operator (under NDA)
Industry
Telecommunications
Period
2022 – present
Status
In production · new version in certification
13
transaction types in production
156+
documented transaction flows
380
automated tests

Product figures. The client's name and the details of its operation are protected by a confidentiality agreement.

The problem

What wasn't working?

At a point of service, every transaction on a phone line starts with the same question: is this person who they say they are? Without a reliable way to answer it, any transaction is an open door to identity fraud.

  • Sales, ownership changes, and SIM replacements are common targets for identity fraud
  • Every transaction changes depending on who shows up: the account holder, a legal representative, or an authorized third party, individual or business
  • The service network mixes company-owned stores and retail partners, each with different rules
  • Case files gather many documents: ID, powers of attorney, company paperwork, proof of payment
  • After four years of growth, the first version had 59 hard-coded step sequences: every rule change meant touching dozens of screens
The solution

What I built

First I built the app that digitized verification and took it to production. Then, with everything learned along the way, I rewrote it completely on a foundation designed to change rules without rebuilding screens.

01

The agents' app

The agent signs in, looks up the customer, captures what the transaction requires, and submits the complete file after a review screen. 13 transaction types in production.

02

Complete biometric verification

Face with liveness detection, fingerprint, face-to-ID-photo matching, and a handwritten digital signature, all from the agent's phone.

03

Automatic ID reading

OCR pulls the name, number, and dates from the ID document and scores how closely its photo matches the captured face. Less typing, fewer errors.

04

Two ways to capture a fingerprint

An external Bluetooth fingerprint reader or, as a fallback, the phone's own camera. If one device fails, service doesn't stop.

05

One engine instead of 59 sequences

In the new version, a single component decides the steps for each transaction based on the customer type, payment method, and who is present. Changing a rule no longer means rebuilding screens.

06

Visibility in production

Built-in error and usage monitoring, to catch a failure before an agent has to report it.

The result

What changed for the client?

The original app remains in production with ongoing maintenance. The new version has been delivered to the client for certification and will replace it.

  • In production since 2022, with five releases in 2026 alone
  • From 59 fixed sequences to a rules engine that handles more than 156 documented transaction flows
  • 380 automated tests covering the business rules of every transaction
  • Memory usage cut from 220 MB to about 120 MB, with 20 consecutive transactions and no crashes
  • The document-reading provider can be switched, and switched back, with a toggle: no new app release required

Technology used

React NativeFlutterDartKotlinRiverpodFirebaseBluetoothOCR

Have a similar project in mind?

Let's start with a free Discovery Session. In 5–8 minutes I'll pin down exactly what your project needs and give you a straight proposal.

Start a Discovery Session

Free · No strings attached · I reply personally