E-Commerce Chatbot

An example of a chatbot built with Microsoft Bot Framework and featuring e-commerce capabilities via:

I presented this bot on API Strat in Boston as an example of a smart app built with cognitive APIs.

Video

If you would like to run it, you would need:

A Moltin subscription with the Adventure Works data (I previously shared scripts to import Adventure Works data into Moltin)
Azure Search service with three indexes - categories, products, and variants. You can find the index definitions here and I will soon share the script that creates indexes and loads the data
Recommendations API endpoint with the FBT (frequently bought together) model trained on historical orders. I will soon share the script that loads the data and trains the model
Trained LUIS model for the intents that require NLU to be recognized. You can import the app that I trained to get a head start

Deploy your bot (I used Azure App Service) and register it with the dev.botframework.com.

Set the following environment variables:

MICROSOFT_APP_ID - you will get it from the dev.botframework.com during registration
MICROSFT_APP_PASSWORD - you will get it from the dev.botframework.com during registration
RECOMMENDATION_API_KEY - your API key to the Recommendations API service from the Microsoft Cognitive Services
RECOMMENDATION_MODEL- you can create multiple recommendation models and this way you can choose which one the bot will use for suggestions
RECOMMENDATION_BUILD - a given model (your product catalog, historical transactions, and business rules) can have multiple recommendation builds and this is how you tell which one the bot will use
SEARCH_APP_NAME - the name of your Azure Search service. The code assumes that you have all three indexes in the same Azure Search resource
SEARCH_API_KEY- your API key to the Azure Search service
LUIS_ENDPOINT - the URL of your published LUIS model (important: microsoft/botframework-sdk#1633)

The shopping cart is currently kept in the bot's memory (session.privateConversationData.cart) and does not sync back to Moltin
Checkout process is not integrated with Moltin

MIT