Sara - the Rasa Demo Bot

🏄 Introduction

The purpose of this repo is to showcase a contextual AI assistant built with the open source Rasa Stack.

Sara is an alpha version and lives in our docs, helping developers getting started with our open source tools. It supports the following user goals:

Understanding the Rasa Stack
Installing the Rasa Stack
Answering some FAQs around the Rasa Stack
Subscribing to the Rasa newsletter
Requesting a call with Rasa's sales team
Handling basic chitchat

You can talk to Sara here and find planned enhancements for Sara in the Project Board

🤖 How to install and run Sara

To install Sara, please clone the repo and run:

cd rasa-demo
pip install -e .

This will install the bot and all of its requirements. Note that it was written in Python 3 so might not work with PY2.

Training CORE and NLU

Train CORE

To train the core model: make train-core (this will take 2h+ and a significant amount of memory to train, if you want to train it faster, try the training command with --augmentation 0). We set to this to reduce training time. If you want to use that, please remove the change from Makefile To train the core model:

make train-core

If you want to train it faster, try the training command with --augmentation 0) as

(modify Makefile)
as
train-core:
        python3 -m rasa_core.train -d domain.yml -s data/core -c policy.yml --debug -o models/dialogue --augmentation 0

Train NLU

To train the NLU model:

make train-nlu

Run models

To run Sara with both these models:

docker run -p 8000:8000 rasa/duckling
make run-cmdline

There are some custom actions that require connections to external services, specifically ActionSubscribeNewsletter and ActionStoreSalesInfo. For these to run you would need to have your own MailChimp newsletter and a Google sheet to connect to.

If you would like to run Sara on your website, follow the instructions here to place the chat widget on your website.

👩‍💻 Overview of the files

data/core/ - contains stories for Rasa Core

data/nlu - contains example NLU training data

demo - contains custom action/api code

domain.yml - the domain file for Core

nlu_tensorflow.yml - the NLU config file

policy.yml - the Core config file

models - the Core models persistence folder, i.e., models/dialogue

├── augmentedmemo-only.yml
├── data
│   ├── core
│   │   ├── canthelp.md
│   │   ├── chitchat.md
│   │   ├── closetheloop.md
│   │   ├── faqs.md
│   │   ├── feedback.md
│   │   ├── get_started.md
│   │   ├── handoff.md
│   │   ├── oos.md
│   │   ├── step3_install_rasa.md
│   │   ├── step4.md
│   │   └── stories.md
│   ├── intent_description_mapping.csv
│   └── nlu
│       └── nlu.md
├── demo
│   ├── actions.py
│   ├── api.py
│   ├── community_events.py
│   ├── config.py
│   ├── gdrive_service.py
│   ├── __init__.py
│   └── __pycache__
│       ├── actions.cpython-36.pyc
│       ├── api.cpython-36.pyc
│       ├── community_events.cpython-36.pyc
│       ├── config.cpython-36.pyc
│       ├── gdrive_service.cpython-36.pyc
│       └── __init__.cpython-36.pyc
├── Dockerfile
├── domain.yml
├── endpoints.yml
├── LICENSE
├── log-rasa-core-training.pdf
├── logs
│   ├── log-rasa-core-training.pdf
│   └── pip-install-error.txt
├── Makefile
├── models
│   ├── dialogue
│   │   ├── domain.json
│   │   ├── domain.yml
│   │   ├── policy_0_KerasPolicy
│   │   │   ├── featurizer.json
│   │   │   ├── keras_model.h5
│   │   │   └── keras_policy.json
│   │   ├── policy_1_AugmentedMemoizationPolicy
│   │   │   ├── featurizer.json
│   │   │   └── memorized_turns.json
│   │   ├── policy_2_TwoStageFallbackPolicy
│   │   │   └── two_stage_fallback_policy.json
│   │   ├── policy_3_FormPolicy
│   │   │   ├── featurizer.json
│   │   │   └── memorized_turns.json
│   │   └── policy_metadata.json
│   └── nlu
│       └── current
│           ├── checkpoint
│           ├── crf_model.pkl
│           ├── entity_synonyms.json
│           ├── intent_classifier_tensorflow_embedding.ckpt.data-00000-of-00001
│           ├── intent_classifier_tensorflow_embedding.ckpt.index
│           ├── intent_classifier_tensorflow_embedding.ckpt.meta
│           ├── intent_classifier_tensorflow_embedding_encoded_all_intents.pkl
│           ├── intent_classifier_tensorflow_embedding_inv_intent_dict.pkl
│           ├── intent_featurizer_count_vectors.pkl
│           ├── metadata.json
│           └── training_data.json
├── nlu_tensorflow.yml
├── pip-install-error.txt
├── policy.yml
├── rasa_core.log
├── rasa_demo.egg-info
│   ├── dependency_links.txt
│   ├── PKG-INFO
│   ├── requires.txt
│   ├── SOURCES.txt
│   └── top_level.txt
├── README.md
├── requirements.txt
├── setup.py
└── travis_gcloud_auth.json.enc

Code Style

To ensure a standardized code style we use the formatter black.

If you want to automatically format your code on every commit, you can use pre-commit. Just install it via pip install pre-commit and execute pre-commit install in the root folder. This will add a hook to the repository, which reformats files on every commit.

If you want to set it up manually, install black via pip install black. To reformat files execute

black .

DrSnowbird/rasa-demo