Skip to main content

 In the world of Intelligent Automation (IA), efficiency and accuracy are paramount. One common challenge involves comparing two property addresses across different systems in use cases like Accounts Payable/Order fulfilment Process/Insurance Claim Processing etc making it a cumbersome and error-prone task.

Consider a use case where we need to compare two property addresses from two different systems.

Example 1:

System 1 Address : 555 Oak Ave, Queens, NY 11370-1234
System 2 Address : 555 OAK AVENUE, QUEENS, NEW YORK 11370-1234

These addresses have the same street name ("Oak Ave") and are in the same city (Queens, NY). The differences are in letter case and the use of full words for city names, but these do not change the core address hence it’s a match.

Exploring further examples reveals that addresses may contain different combinations of special characters and do not necessarily require an exact match in subsequent systems. The comparison can be performed using Address Line 1, Line 2, State, City, and Zip Code.

Example 2 :

System 1 Address : "555 Oak Ave, Queens, NY 11370-1234"
System 2 Address : "555 Oak AVENUE, queens, NY 11370-1234"

Both addresses have the same street name ("Oak Ave") and are in the same city (Queens, NY). The differences are in the letter case and the use of full words for "AVENUE," but the core address remains the same.


We can achieve this using traditional programming languages and then integrate with A360 or build a very complex logic within A360 platform, involving the lookup of abbreviations via APIs, databases, or Excel sheets for comparisons. However, this approach comes with substantial risks related to outcomes and maintenance.

To solve this challenge, we will use LLMs (Large Language Models). Using a single prompt, we can achieve more accurate outputs. This demo will utilize the OpenAI Chat Completions API.
 

OpenAI Account Setup

 

  1. Signup for OpenAI Account, As of today we get 5💸 free credit to playaround.
  2. Sign in to your account
  3. Navigate to API keys >> click on “Create new secret key
  4. Copy the secret key, we will use it at a later point in this demo.

Building the Automation

 

  1. Login to Community or Enterprise CR
  2. Navigate to Automations >> Private Repository >> Create a new bot
  3. Use the “Prompt” package for keying in the 2 addresses i.e System 1 Address and System 2 Address. Create 2 new variables as shown below for holding these values, i.e sSystem_1_Address & sSystem_2_Address respectively.

     

  4. Use “Generative AI” package, which supports various Large Language Models (LLMs), for this demo, we willuse OpenAI.
  5. Use Generative AI > OpenAI, drag Authenticate action and provide input parameters as below,
    1. Provide APIKey generated from the above step dOpenAI Account Setup], for simplicity of the demo, we will be passing the KEY directly as an insecure string.
    2. Provide session name accordingly, here I am leaving it as the Default local session.

       

  6. Use Generative AI > OpenAI, drag Chat AI action and provide input parameters as below,
    1. Leave session name as Default
    2. Choose a model based on your preference, go for gpt-3.5-turbo if you are on a free plan or use GPT-4 if it's a paid one.
    3. Provide PROMPT, Max tokens and Temperature
      • Note : The temperature value influences the randomness of the response. As the temperature approaches zero, it makes the response more focused and deterministic. The higher the value, the more random the response.
    4. Pass the 2 variables which hold 2 addresses from the user in the PROMPT, as shown below.
    5. Create a new Output variable to store the API response.
    6. OPTIONAL : To manage the optional parameters, click Show more options and select Yes. If you select Yes, you can add other parameters such as:
      • Top P
      • Stop
      • Presence Penalty
      • Frequency Penalty
      • User

         

  7. Use Message Box action to display the output,

     

  8. Completed Automation will look like as shown below:

    9. Save the changes to the bot.

Execution

 

  1. Key in System 1 & 2 Address and click Ok.

     

  2. Output

     

  1. Retrieve the JSON output generated by the OpenAI Chat Conversations. Extract the relevant information, such as the result, reason, and confidence level, for further processing

          OR

          Re-route these results to users using Automation Copilot via forms for the items where we have low confidence for human review.

Conclusion

 

This tutorial demonstrates techniques for utilizing Large Language Models (LLMs) to effectively address data-related challenges and enhance the capabilities of intelligent automation solutions.

Now, its your time to think about similar data-related challenges you have faced in your automations and how you can utilize LLMs to solve them.

Be the first to reply!

Reply