Survey Analyzer

A quick demo web application using Azure's AI services (Document Intelligence and Language for Sentiment Analysis) that takes a specific feedback template (whether by PDF or image), extracts text into fields such as name, gender, email, and feedback using a custom Document Intelligence model then applies sentiment analysis on a survey's feedback

Technologies used: React, Flask, Docker, Terraform, and Azure

• Getting Started
  • Steps
    • Local
    • Azure VM
    • Document Intelligence Model Training
    • Testing
• Wrap Up

Getting Started

• Make sure the following are installed
  • Docker
  • Docker Compose
  • Terraform
  • Azure CLI
• Login with your Azure account az login

Steps

The default setup deploys the application to an Azure virtual machine with an Nginx web server configured with SSL automatically. If you'll use the app locally you won't need to deploy a virtual machine, follow the steps below and choose either the Local steps or Azure VM steps

Local

• Rename virtual-machine.tf to virtual-machine.tf.disable
• Comment vm_dns_name and vm_public_ip code blocks in the outputs.tf terraform file
• Deploy the infrastructure terraform init && terraform apply
• Follow the Document Intelligence Model Training steps below
  • If you named the model something else than survey-model, be sure to change DOCUMENT_INTELLIGENCE_MODEL_ID accordingly in the env-files-setup.sh script
• Update the .env files using modify-env-vars.sh script which takes Terraform outputs and adds them in the .env file of both Frontend and Backend
• Deploy the app docker compose build && docker compose up
• Go to http://localhost:3000 and upload a survey

Azure VM

• Deploy the infrastructure terraform init && terraform apply -var public_key_path=~/.ssh/key-name.pub
• Follow the Document Intelligence Model Training steps below
• Update the .env files using modify-env-vars.sh script which takes Terraform outputs and adds them in the .env file of both Frontend and Backend
• In the repo directory, copy the docker-compose.yml file and backend and frontend directories to the server

  # You can get the public IP of the VM whether from Azure or using the command below
  terraform output -state $(readlink -f $(dirname .))/terraform/terraform.tfstate -raw vm_public_ip
  
  # You may need to include path to private key
  scp -r docker-compose.yml backend/ frontend/ ubuntu@$public_ip:.
  

• ssh ubuntu@$vm_public_ip then docker compose build && docker compose up -d to deploy the app
• In Azure portal, find the DNS name in the VM, put it in the browser and upload a survey
  • Alternatively you can execute terraform output -state $(readlink -f $(dirname .))/terraform/terraform.tfstate -raw vm_dns_name to get the DNS

Document Intelligence Model Training

1. Go to Azure's Document Intelligence studio and create a new project

   

2. You can name the project anything you like, make sure to target the correct resource group, storage account and blob container like shown below

   

3. Run layout to scan all the available text, and create 4 fields, then highlight each text and add it to the correct field

   • Note: keys such as Feedback are case sensitive, the app won't work if its written feedback instead

   

4. Repeat on the rest of the PDFs then finally click Train on the top right corner to train the model, be sure to name it survey-model

   • Note: you can name the model anything else you like, just be sure to change MODEL_ID accordingly in the backend's .env file

   

Testing

After the app has been successfully deployed, you can use the survey forms in the terraform directory or create your own by using this template (Canva account will be needed to edit) Upload the custom survey file then click on analyze and you're done

Wrap Up

Be sure to not forget to destroy your Terraform resources after you're done. If you have any feedback or any issue don't hesitate to let me know!