CS01 - Virtual Assistant for UofG's External Relations

This software engineering project is to create a virtual assistant / chatbot that will help answer the increasing volume of application and course enquiries at the University of Glasgow's External Relations Directorate.

Our application consists of:

an interface for users to interact with the chatbot (Flask + socketio)
connected to a Natural Language Understanding unit (RASA)
which makes queries to a free-text search engine (Elasticsearch)

Launching with Docker
Sample questions to ask the bot
Contribution guide (includes codebase structure information)
Set up dependencies manually
- Launching from manual set-up

Set up

Download a release of UGenie, which should include a data archive and a model archive
- models.zip contains a trained instance of the bot
  - it should be decompressed in chat-service/agent-data/ to create the path chat-service/agent-data/models
- data.zip contains the client-provided university data
  - it should be decompressed in elastic-db/ to create the path elastic-db/data

Launching with Docker

This is easier for direct interaction with the bot and dependency setup.

docker-compose up --build

This should populate the database when launching. The bot should then be running at localhost:5000

Manuel dependency setup

How to get yourself set up with dependencies manually, for development:

pip install rasa_core
pip install rasa_nlu[spacy]
python -m spacy download en
pip install -r requirements.txt # socket.io + elasticsearch wrapper + xlrd

N.B. Python should be 3.6+. RASA installation is done in steps because it is a heavy package, and linking the spaCy english model only works this way.

Elasticsearch also needs to be installed and ran as a background process:

with Docker

docker-compose up elastic

Installing manually

Installation varies quite a bit depending on OS, please look through the official Elastic reference for instructions!

Launching after setup

make populate # populate elastic with the data from data.zip
make train # not necessary if the models.zip was downloaded and nothing has been changed
make run

The bot should then be running at localhost:5000
N.B.: if chat logs become excessive, run make clean f=[filename] to zip them up!

Interacting with the bot

Here are some questions you can ask UGenie: