How to Run In Docker:

Flask app is setup to port 8080 and broadcast on 0.0.0.0 due to dockerization

Build it
    docker build -t aggregator .

Run it
    docker run -p 8080:8080 aggregator

Test it
    test_curl_with_full_data.sh

How to run in CLI

cat input_data.json | python data_aggregator_cli.py currency country
cat input_data.json | python data_aggregator_cli.py currency country city

How to run without docker

python aggregator_flask_app.py

Flask need authentication (see test_curl_with_full_data.sh )

Or in the source :)

Future Improvements

Application could have register endpoint for registering tokens and save it locally or to database

Create minimal database to contain data and result for future retrieving of the data. Alternatively grabs results from filesystem under requests ip address (substitution for user)

Approach

During the development I have applied TDD in order to ensure and fast test of desired outputs. In the test are classes three classes. Two of those are to test classes described bellow and third as integration test that is bound to input_data.json

In file datahandler I have created two classes:

OriginData
DataHandler

Each of those classes has a responsibility, going bottom-up

OriginData

Is class to elemental data such as: { "country": "US", "city": "Boston", "currency": "USD", "amount": 100 } inside a class where is stored, class has a function to handle loading input data and transformation to desired state according to input keys parameters

DataHandler

Is class that contains list of the OriginData, handles loading the data from multiple inputs and feeding it in correct state into the OriginData (not validated hardly enough to be called exactly robust, but it is what it is).

Also handles the aggregation of the data into according to the keys desired from input.

This approach was chosen to have multiple classes that has single responsibility of handling the data. It was used as this is open to future implementations and changes to more seamlessly integrate into system if in the future will be request to handle data in different way and to still contain old functionality

Conclusion

In this task I have implemented a functionality of data transformation and aggregation accordingly to keys in the input json. There is a CLI version of the application that handles file input from stdin and keys for transformation. As there is no exact specification for number of aggregation keys I have decided that there should be at least two. Another working solution is web application written in Flask with only single POST endpoint /api/v1/parse-me/ . Try/see test_curl_with_full_data.sh. This application is also dockerizable

Another approach to handle the data transformation was with using pandas library, where instead of classes will be one transformation function to achieve same result and would be faster in implementation than painfully transform the nested dicts in desired conditions. Also app is lightweight and only dependency is Flask.