Use Azure OpenAI and LangChain to categorize a document by combining input document with provided template documents

In this tutorial, we will learn how to categorize the documents using Azure OpenAI and LangChain.

As part of my daily responsibilities, I am tasked with reviewing change request (CR) documents submitted by my team. These CR documents come in two types: Normal and Standard. The primary distinction between these two types is that the Standard version is a simplified variant of the normal CR document. I have also made templates available for both document types to assist my team in adhering to a standardized format when submitting CRs. However, there are instances when the templates are not followed, leading to the need for request rejection. To facilitate both my team’s and I reviewing the CR document, I have developed an application for document review and categorization.

CR Templates

I have created CR template documents for both the standard and normal CR, which I will input into LangChain.

Standard CR (CR_Template_Standard.txt)

THIS IS STANDARD CR TEMPLATE DOCUMENT — START

Change Description
==================

Description:
<<jira id>>
Man hours Saved annually: <<man hours annually>>

Business Background :
<<description>>

Justification
=============
Business Justification:
<<business justification>>

Implementation Plan
===================
Deployment of this CR will be done on <<deployment date>>
E.g. Tuesday 05-Sep-2023
Deployment estimate timing: <<date with time in SGT >> e.g.10:30 AM SGT — 11:00 AM SGT
Validation/Sanity: <<date with time in SGT from and to >> e.g. 05-Sep-2023 11:01 AM SGT — 07-Sep-2023 07:00 PM SGT
New / Updated Recipe to be deployed to production:
New / Updated Recipes:
<<list of new/updated recipes>>

Lookup Table:
<<Lookup table name>>

Message Template:
<<message template name>>
Please mention other attributes if they are applicable
Who will perform the activities?
Deployment: <<name>>
Developer : <<name>>
List of applications impacted:
<<applications>>

Backout Plan…