Cognitive Merchandiser

The Cognitive Merchandiser is a reference application built on the following Premise:

Hypothesis: Uncaught Intents and Unstructured Web-experiences

While the former is an issue with Intents not being captured properly and responed to, the latter is the solution for the problem.

Over $4.3 Trillion dollars worth of merchandise was left in checkout pages in 2014 ( http://www.businessinsider.com/heres-how-retailers-can-reduce-shopping-cart-abandonment-and-recoup-billions-of-dollars-in-lost-sales-2014-4)

Top reasons are shipping costs but also I was just browsing The Retail experience wasn't able to complete a sale because of two reasons that I am Hypothesising

Out of touch experience creation processes.

• User experience as well as persona's are getting built much prior to the actual application development.
• Once created , Designers draw based on these and leave.
• Implemented UX Personas are captured generically, sample this quote:

Each persona represents a significant portion of people in the real world and enables the designer to focus on a manageable and memorable cast of characters, instead of focusing on thousands of individuals. Personas aid designers to create different designs for different kinds of people and to design for a specific somebody, rather than a generic everybody (https://www.smashingmagazine.com/2014/08/a-closer-look-at-personas-part-1/)

Overload of Digital Marketing tools

The solution

A Cognitive layer is proposed on top of everything else, it caters to *One* and can operate on multiple concepts simultaneously and can scale at will.

The Cognitive App is considered a succcess if:

• It takes in a plainly worded but Intent oriented persona and returns a cognitive set of characteristics which will be used as a training set.
• The visual recognition API learns Catalog and Maps intents ( A Retailer sample in a catalog.js is checked in data folder)
• The Relationship Extractor ML API retreives and matches information between the Persona and the Catalog.
• User supplies a search result say White Jacket, a NLP system undertands that White is a color and Jacket is in the Catalog matched to the Alpha Male
• System matches and returns all white jackets as a page on the Site
• This page was not defined anywhere.. The page components are automatically merged in and user experience components chosen accordingly.
• YMMV: The App should then attempt to Title this page and fetch potentially a related article.

Key Technologies used

• watson developer cloud
• webpack
• ReactJS
• Redux

Getting Started

• Node, NPM and Mongo are the only requirements right now.
• npm install
• npm run dev or npm run build
• Open http://localhost:3000/Persona
• Fill the Brand brief and submit ( See Issues)

Status

What works

Everything is wired up using the NodeJS SDK for Watson Cloud. Actions, Reducers are all wired in and respond to changees.

Boiler plate credits

React-Go for the Base React, Redux and Express Architecture: https://github.com/reactGo/reactGo

Roadmap

Re-inforcement learning using Pavlov https://github.com/nathanEpstein/pavlov.js

Built With

• css
• gcc-machine-description
• javascript
• react
• redux