The heroku link for this app is http://rs-challenge.herokuapp.com

Development setup

To install the gems

bundle install

To setup the database

bundle exec rake db:setup

To run database migrations

bundle exec rake db:migrate

To import CSV

bundle exec rake app:import_csv_data

To run all the test cases

bundle exec rspec spec 

To run the Rails server

bundle exec rails s 

To drop the database

bundle exec rake db:drop

Importing Records From CSV files

• The records are divided into chunks rather than loading the whole file and process them in batch using SmarterCSV and threads. The helper for the rake file is in ROOT/lib directory

UI of Application

• The appplication would root a page where 500 users from database are displayed and an algorithm is chosen from the dropdown which would route to other page where the recommended items are displayed as table

Database Scaling

• Created Indexes on Foreign Keys which speeded up the whole process 4X-5X times
• Active Record Caching - https://github.com/Shopify/identity_cache (memchached based which also keep track of modified objects in cache) which I think is another way to scale the database to a great extent. Didn't implement in this project because I need to change the schema as identity_cache doesnt support :through. So, I just tested the gem's functionality by creating memchached server using Dolli
• Can use NoSQL databases (Distributed) like Mongodb, Elasticsearch which would greatly improve the speed.

OOPS design

• Used Strategy Design pattern to design the recommender algorthm as they have same interface and the algorithm can be chosen at runtime based on the option user chooses in the view
• The algorithms are in app/services directory
• Furthur, regarding the design all the algorithm classes can extend the base class and would have raised an error if they didn't implement the base class method recommend. But to keep the design simple I didnt do that

Some Other insights

• All the algorithms used are naive and can use algorithms like collobarative filtering and other statistics methods for accuracy
• I placed all the activerecord queries in algorithms as scopes in model to be reusable in future and readable, although it is an overkill at this stage of project
• Can have page caching to save time of rendering the page