Nova AI

See a Loom Demo of NovaAI here

Try out Nova at:

http://nova-ai.net/ (Udpate: no longer active)

Overview

Nova is an application that uses conversational AI for language learning to help users gain fluency in their target language (work in progress).

Users can chat with Nova, practice specific scenarios, and track their learning progress.

Features:

• A language learning website implemented Flask
• Chats with a conversational AI powered by GPT-3
• Grammar correction and other learning features

Usage

Installation

The package uses python3.9.

First clone the repo:

git clone git@github.com:ConvAF/NovaAI.git
cd chatbot

It is recommended to also set up and activate a virtual environment, for example using

pip install virtualenv
virtualenv venv
source venv/bin/activate

Then, install the application and dependencies as a package in development mode using

make install

which is equivalent to running

pip install -e .

which installs the application as a package in developer (local edit) mode.

Configuration

Create a file .env at the project root (if it doesn't exist yet) and specify the following variables before running the application:

Several variables are stored in the .env file and should be configured before running the application:

• OPENAI_API_KEY: The api key for OpenAi.
• OPENAI_ENGINE: Which GPT-3 engine to use. For testing, a simpler engine is used. Can be one of: ada (cheapest), babbage, curie, davinci (most expensive, most powerful)

Running the application

Before running the application, initialize the user database

flask init-db

Start the application from the root project directory using

flask run

If the application is installed, you can also start it from anywhere using

export FLASK_APP=chatbot
flask run

Deployment

Deploying on AWS EC2

This describes how to deploy the app on an AWS EC2 instance using docker.

First clone the repo:

git clone git@github.com:ConvAF/NovaAI.git
cd chatbot

Then build the image using

docker build . --tag chatbot

To run the app (and keep running in background), start the container as

docker run -d chatbot:latest

Be sure that the correct port (80 by default) is exposed on your instance such that the application is accessible from the outside (see notes below).

Developer Notes

Useful commands

• flask run: Run the application.
• flask init-db: Initialize the database.
• flask shell: Start an interactive Python shell in the application context, with an app instance imported

Testing

Running Unit Tests

You can run tests using

pytest

Add the -v flag to show the result for each test function.

To measure the code coverage of the tests, use the coverage command:

coverage run -m pytest

You can view the coverage report using

coverage report
# or
coverage html

Testing in the Flask shell

The flask application can also be tested from the shell.

Start the shell with

flask shell

In the shell, you can test endpoints like so:

>>> client = app.test_client()
>>> client.get('/')
<WrapperTestResponse streamed [200 OK]>
>>> client.post(
      '/auth/register',
      data={'username': username, 'password': password}
    )

Load testing with locust

To test the application under user traffic, do

cd tests/
locust

which will open a dashboard at http://0.0.0.0:8089/. There you can simulate users retrieving different endpoints. More complex fuctionality can be added in tests/locustfile.py.

Profiling with py-spy

To profile the app, first get the process id (pid) of the running flask app, for example using

ps | grep flask

Then you can record the app's activity using

py-spy record -o profile.svg --pid <pid>

You can interact with the app for a while (or run a locust load test), after a while stop the py-spy process, which will save a flame graph of the apps activity.

Notes on manual deployment on AWS EC 2

The following outlines some steps if not using docker as described above

Installing pytorch

We encountered a bug when installing pytorch on an AWS EC2 instance, where the installation with pip did not finish. To circumvent this, use

pip install --no-cache-dir torch

or, for all requirements,

pip install -e . --no-cache-dir

Serving the app

waitress-serve --call 'chatbot:create_app'

Ports on AWS EC2

waitress serves the app on port 8080. To redirect to port 80, use

sudo iptables -t nat -I PREROUTING -p tcp --dport 80 -j REDIRECT --to-ports 8080

Docker

Docker image is specified in the docker file.

Build the image using

docker build . --tag chatbot

Start the container (and keep running in background)

docker run -dp 80:80 chatbot:latest

For testing, run the container and delete after user with

docker run -p 80:80 --rm chatbot:latest

(ports are not necessary on the server, but on mac you have to specify the ports 8080:80 because 80 is already in use).

or in interactive mode to inspect the contents:

docker run -it --rm chatbot:latest bash