Payments / SWIFT · GTI · Dec 2025 - Feb 2026
Structured address extraction for ISO 20022 payments
Converts free-text addresses in SWIFT payment messages into ISO 20022 structured postal-address components - the migration every bank faced ahead of the structured-address mandate.
At a glance
- Modelled the full 16-field ISO 20022 schema against the payment-message specification, then bootstrapped a seed corpus of around 8,000 auto-labelled addresses with a validation pass before human review.
- Benchmarked spaCy pipelines against a multilingual transformer across three corpus generations up to 41,000 samples, and shipped an ensemble extractor.
- Built an augmentation toolkit that beat naive scaling, and a companion country matcher covering 2,532 name variations across 248 country codes in six languages.
Stack
PythonspaCyHuggingFace TransformersXLM-RoBERTaPyTorchseqevalLabel StudioDocker
Repo and demo
Client work, code is private. The architecture and the evaluation method are covered above, and I'm happy to walk through the code in an interview.