Address App

TLDR: Mapping free-text address -> Structured fields using machine learning.

This project has an api and application to map free-text street addresses to structured Australian GNAF fields.

It borrows heavily from the excellent address-net project and uses a pretrained Tensorflow model to parse and segment the free text address.

Working examples

Application: https://address-app.infocruncher.com/

Api: https://address-api.infocruncher.com/

Project Structure

App

The /client folder is a Terraform managed S3 website app hosted at address-app.infocruncher.com that uses the api endpoint.

See /client/Makefile for more information around deploying the app to AWS.

Api

The project root is an AWS SAM managed API address-api.infocruncher.com consisting of an API Gateway with Lambda backend.

The actual app that is deployed to Lambda is Dockerised and can be found in /app

See /Makefile for more information around deploying the api to AWS.

Example Api response given a POST request with Unit 18/14-18 Flood St, Bondi, NSW 2026:

{
  "address": "Unit 18/14-18 Flood St, Bondi, NSW 2026",
  "result": {
    "flat_type": "UNIT",
    "flat_number": "18",
    "number_first": "14",
    "number_last": "18",
    "street_name": "FLOOD",
    "street_type": "STREET",
    "locality_name": "BONDI",
    "state": "NEW SOUTH WALES",
    "postcode": "2026"
  },
  "handler_time": "0:00:01.180911",
  "runtime_time": "0:04:24.914928",
  "model_dir": "/opt/ml/model/pretrained",
  "version": "0.1.11"
}