Deploy your apps to a supercloud in a few clicks

This Engineering Education program is supported by Section. Instantly deploy your GitHub apps, Docker containers or K8s namespaces to a supercloud.

Try It For Free

How to build a Q&A web application using Python and Anvil

February 21, 2022

Question-Answering web apps have been around for a while now. A good example is the use of bots on websites where a user asks a question, and an answer is generated automatically.

In this build, we will show our readers how to build a Q&A app using one of the pre-trained models on HuggingFace and Anvil.

Anvil is an open-source web interface that allows a developer to spin up and deploy a web app using Python. It is quite similar to the popular Gradio web application used by developers. We will leverage Anvil to build our front-end.


To follow along, you need to have:

  • An understanding of Natural Language Processing.
  • An understanding of Machine Learning modeling.
  • Google Colab or Jupyter Notebook.


Installing and importing dependencies

The first dependency that we will install is PyTorch.

!pip3 install torch==1.10.2+cu113 torchvision==0.11.3+cu113 torchaudio==0.10.2+cu113 -f

The next dependency that we will install is Transformers. It’s the NLP library that will allow us to use the model from HuggingFace.

!pip install transformers

Those are the two main dependencies that we will use in this tutorial. Let’s go ahead and import the tokenizer, pipeline, and model from transformers.

from transformers import AutoModelForQuestionAnswering, AutoTokenizer, pipeline
  • The NLP model uses the AutoTokenizer to convert natural language into tokens.
  • The AutoModelForQuestionAnswering allows us to load our model.
  • The pipeline allows us to use the model.

We can now go ahead and set up our model.

Setup the deep learning NLP model

Let’s begin by creating the Q&A pipeline.

model = 'deepset/roberta-base-squad2'

nlp_pipeline = pipeline('question-answering', model= model, tokenizer= model)

When you run the code above, it will download the model to your notebook.

You don’t have to use the same model we’ve used in this tutorial. We chose the deepset/roberta-base-squad2 as it is the most popular. But, there are many Question Answering models on HuggingFace. Feel free to experiment with any other too.

To use this model, we’ll need to pass in a dictionary. To this dictionary, we need to pass in two things. A question and some context.

We create a variable called random_text to hold some text randomly fetched from Wikipedia. We will pass this variable to the context.

random_text = """
<To avoid plagiarism flags, please refer to the linked Google Colab URL to access the random text gotten from Wikipedia>

QA_set = {
            'question':'What is machine learning?',
            'context': random_text

Let’s go ahead and test this out in our pipeline.



{'answer': 'computer algorithms that can improve automatically through experience and by the use of data',
 'end': 131,
 'score': 0.06677386909723282,
 'start': 39}

Impressive right? Let’s pass on another question.

QA_set = {
            'question':'Where are machine learning algorithms used?',
            'context': random_text


{'answer': 'medicine, email filtering, speech recognition, and computer vision',
 'end': 509,
 'score': 0.27500787377357483,
 'start': 443}

The model seems to be giving us the answers by simply fetching context from the paragraph. This is good. We can take this a step further. We can build a web interface that enables users to interact with the model easily. This is where Anvil comes in handy.

Integrating the model with Anvil app

Anvil is a platform for building full-stack web apps entirely in Python. It makes it easy to build and deploy full-stack machine learning apps. It is easy-to-use and open-source. We will use the framework for this build.

To use it, head on to their website and click on the start building button on the top-right corner of the webpage. Of course, you’ll need to sign up to access the app. If you already have your logins, you can go ahead and log in. If you don’t, you can sign up then log in. It is free.

On successful login, create a new blank app. You’ll be prompted to choose the theme of your choice. We selected the Material Design theme for this build.

If you’ve ever used WordPress before with its drag and drop feature, building the front-end is quite similar.

Here’s the design we come up with:


Now that we have our design ready, we need to head back to our Google Colab and connect to Anvil from there.

Connecting to Anvil

We begin by installing the main dependency called anvil-uplink. It allows us to connect our Colab to the Anvil app.

!pip install anvil-uplink

As with the previous installs, we import it into our Colab.

import anvil.server

The next step involves connecting to the Anvil server.

anvil.server.connect('<Add your Anvil code here>')

The screenshot below shows you where you can get the anvil code on the app.

Anvil code

Click on the Uplink option. A pop-up will appear as shown below:

Anvil server uplink

Finally, you need to enable the Enable the Anvil Server Uplink for this app button to get the code.

When you run the code, you should see the message below showing that you have successfully connected to the server.

Connecting to wss://
Anvil websocket open
Connected to "Default environment" as SERVER

Now, we need to set up a callable function that the Anvil app can call to talk to our Google Colab. Besides, our Colab will use the function to connect to the NLP model.

Setting up a callable function

We begin by defining a decorator which tells the server that this is a callable function.


Next, we’ll define the function to connect to our NLP model.

def question_answer(question_text, context_text):
    QA_set = {
            'context': context_text

Let’s run it through the NLP pipeline and store the results inside a variable called results.

results = nlp(QA_set)

Finally, we return the answer. Keep in mind that the results are a dictionary with answer, end, score, and start. You can confirm that it returns a dictionary using type(results).

return results['answer']

On our Anvil app, on the code section, add the following function:

 def primary_color_1_click(self, **event_args):
    input_text = self.input_text.text
    question_text = self.question_text.text

    result ='question_answer', question_text, input_text)

    self.answer_text.text = result

This connects our model to the elements we created on the Anvil app.

Testing the app

On the Anvil app, click the Run button. Pick any passage and paste it onto the input area. You can ask any question you would like the model to return an answer. Finally, click on the ASK A QUESTION button.

Here are our results:


We have successfully built a Q&A web application.

Please find the complete code for this tutorial here. You can also access the Anvil Q&A web application here


That’s our Q&A web application built from scratch. It is relatively straightforward.

If you would like to deploy the web application to a wider audience, you can use the Publish this app option on the top-right corner of the Anvil app. It will generate a private link that you can show to your users or embed on your website.

Happy coding!


Peer Review Contributions by: Willies Ogola