It is using Elasticsearch as the (only) persistence layer and more specifically the ActiveRecord Pattern.
Unfortunately this means losing some functionality of Elasticsearch::Model like import tasks, pagination etc.

• Custom rake task for importing data from .json
• Custom pagination since neither kaminari nor will_paginate works
• Score using Wilson lower bound modified for a 5-star rating system
• 5-star rating using Raty and ajax to recalculate score instantly
• 4 different sorting methods depending if there is a query or not
  • Randomly to improve juice discoverability
  • By the custom score calculated using Wilson lower bound
  • By elastic's relevance
  • By combining the two above plus some extra boosts & factors
• Auto complete juice titles with fuzziness using jquery-ui-autocomplete module
• Search through titles, ingredients & tags using the english analyzer
• Filter by juice color
• Filter by ingredient using aggregations in the sidebar
• Combine juice color, ingredient filter, query and sort at the same time
• Use sessions to uniquely seed the random sorting for each user, keep track of his votes and highlight juice ingredients & tags that matched the query
• Sessions auto-expire after a few minutes of not using the site but highlighting also resets when the juice listing changes (visiting any other page than a juice view)
• Sessions use redis for storage in order for the highlighting feature to work, since it was exceeding the 4kb limit of cookies.
• Custom analyzer using my own token filter plugin to singularize and only keep whitelisted ingredients in a subfield
• Identify juice color by using opencv and scikit-learn's k-means algorithm to find the main color in the picture and then calculate the color difference between juice colors using delta e equations (not yet integrated)

• Integrate opencv in an admin panel that requires confirmation before applying found color
• Design not responsive enough for smaller screens and tablets
• Testing
• More...