Skip to content
techgirl.nyc

SWAT Mobility · DHL Turkey

Making messy addresses usable.

I built a language-processing pipeline that turns DHL Turkey’s delivery addresses into structured information for routing systems.

Interactive — click the controls to run it
02 / Read the addressSample input
An unstructured delivery address

Atatürk Cd. No: 12
Kadıköy, İstanbul

Street
—
Building
—
District / city
—

People write addresses. Routing systems need fields.

Illustrative extraction, not a live model or customer address.

Delivery addresses arrive as free text, with street names, building details, districts, and cities written together. Routing systems need those parts separated into consistent fields. At SWAT Mobility, I built a pipeline to do that for DHL Turkey.

I adapted two pretrained language models, BERT and RoBERTa, to recognize and extract the address fields, reaching 98% extraction accuracy. I deployed the pipeline with automated tests and deployment checks so it could be used by downstream routing systems.

98%
accuracy extracting address fields
Built with
BERT / RoBERTaNERCI/CD