Features of the bot:
- Assists the customer in making a purchase
- Offers product recommendations
- Helps the customer in making a return, exchange or refund
- Assists the customer with credit card related issues
- Takes feedback from customer.
(This model was one of the top 26 models in Flipkart Grid 3.0 - Software development track)
To install , please clone the repo and run:
cd rasa-demo
make installThis will install the bot and all of its requirements. Note that this bot should be used with python 3.6 or 3.7.
Use rasa train to train a model (this will take a significant amount of memory to train,
if you want to train it faster, try the training command with
--augmentation 0).
Then, to run, first set up your action server in one terminal window:
rasa run actions --actions actions.actionsIn another window, run the bot:
docker run -p 8000:8000 rasa/duckling
rasa shell --debugNote that --debug mode will produce a lot of output meant to help you understand how the bot is working
under the hood. To simply talk to the bot, you can remove this flag.
If you would like to run the bot on your website, follow the instructions here to place the chat widget on your website.
After doing a rasa train, run the command:
rasa test nlu -u test/test_data.json --model models
rasa test core --stories test/test_stories.mddata/core/ - contains stories
data/nlu - contains NLU training data
actions - contains custom action code
domain.yml - the domain file, including bot response templates
config.yml - training configurations for the NLU pipeline and policy ensemble
To install requirements for development, run:
make install-devTo run unit tests for custom actions:
make test-actions
To ensure proper database cleanup during testing, you will need to include a connection URL for your tracker store database in your .env file e.g.
TRACKER_DB_URL=postgresql:///tracker
This is not necessary for running the actions.