Astound Conversation Workflow Execution Engine

Local development

• Install

  • Docker machine on mac
  • AWS CLI
  • NODE/NPM

• python_utils is a utility submodule for this Repo, to initialize:

  • Inside PyCharm, right click python_utils and Mark Directory as -> Sources root then execute the following ::

    git submodule init git submodule update

• A login script to access Astound docker image repository, ~/bin/ecr-login.sh::

  #!/bin/bash cmd=$(aws ecr get-login | sed 's/-e none //') eval $cmd

• A config file to specify aws reagion, ~/.aws/config ::

  [default] region = us-east-1

• A credential file to store aws ecr login keys, ~/.aws/credentials ::

  [ecr] aws_access_key_id=ASK_ADMIN_FOR_THIS aws_secret_access_key=ASK_ADMIN_FOR_THIS

  Note: An alternative to above two steps is to use aws configure

• pycharm config

  Add a remote interpreter using docker-compose.local.yml

    | Settings -> Project:Illusionist -> Interpreter -> Add -> Docker Compose
    | Select Server as Docker and docker-compose.local.yml as configuration file
  

  Add a configuration using the docker compose interpreter

    | Select the debugger dropdown and chose to edit configurations
    | *Script path*: /usr/local/bin/flask
    | *Parameters*: run -h 0.0.0.0 -p 3000 --no-reload
    | *Environment variables*: PYTHONUNBUFFERED=1;CONFIG_ENV=local
  

• Install UI dependencies ::

  cd ui && npm install

• start docker compose in pycharm, then access a workflow, e.g. 'adidas_luke' ::

  http://localhost:2000/simulator/adidas_luke

Swagger

Swagger expose for api access endpoint at http://<host>:3000/spec

Swagger ui is served seperately. To run it in local, download the swagger ui docker and run on port 3500 ::

docker pull swaggerapi/swagger-ui
docker run -p 3500:8080 -e API_URL=http://localhost:3000/spec swaggerapi/swagger-ui

Documentation

Automated documentation build with Sphinx. In ./docs dir, build html page. ::

  cd doc
  make html

Document can be found in http:///illusionist/docs for example: http://localhost:3000/illusionist/docs

System Design Documents

For detailed design doc, go to Link to Design Document. Here are some highlights:

Bot Designer

Bot Manager

NLP analysis of bot answer performance