About

RAG/LLM supported online migration counseling service & improved Integreat search engine. It integrates as a chat service into the Integreat App and presents requests in a Zammad to counselors. The solution aims to be privacy friendly by not using any third party LLM services.

This project is currently in a research and development phase. The code created for this repo aims to be compatible for future integration into the Integreat CMS. For the time being the code is separated for faster iteration and testing.

Major issues that have to be addressed:

• Support for low ressource languages
• Code mixing
• Language detection
• Translations

Start Project

1. Install a virtual environment and activate it

   python3 -m venv .venv
   source .venv/bin/activate
   

2. Install all dependencies

   pip install .
   

3. Run the server:

   cd integreat_chat
   python3 manage.py migrate
   python3 manage.py runserver
   

Configuration

Back End

• Deploy as normal Django application. No database is needed.

Zammad Integration

To integrate Zammad, the following configuration has to be set:

• Webhook to https://integreat-cms.example.com/api/v3/webhook/zammad/?token=$REGION_TOKEN
• Create a group integreat-chat. All new issues will be assigned to this group.
  • Set an e-mail address for the group
  • Disable the signature
• Create a user for the Integreat CMS
  • Set the user e-mail to [email protected]
  • Assign the user to the integreat-chat group.
  • Grant Agent and Customer roles.
• Ticket attributes (Admin -> Objects -> Ticket):
  • Name: automatic_answers, Display: Automatic Chatbot Answers, Format: Boolean Field, True as default value
  • Name: initial_response_sent, Display: Initial response sent, Format: Boolean field, False as default value
• Create an access token for the Integreat CMS user with permission Agent tickets
• Trigger for webhook:
  • Conditions: Action is updated, Subject contains not "automatically generated message"
  • Execute: webhook configured above
• Auto response for new tickets in each language, exmple is EN
  • Conditions: State is new, Action is updated, Subject contains not "automatically generated message", Title contains [EN], Initial response is no
  • Execute: Note, visibility public, Subject "automatically generated message", Initial response sent yes, add a suitable message in the body.
    • Example message: Welcome to Integreat Chat $REGION_NAME in $LANGUAGE. Our team responds on weekdays, while our chatbot provides summary answers from linked articles, which you should use to verify important information.
• Scheduler to delete old tickets:
  • Run once a week
  • Conditions: state is closed, Last contact before (relative) 6 months
  • Action: delete
  • Disable Notifications: yes