Using Off-Platform AI to Read Documents and Create Records in Salesforce

This post was written by Kicksaw CTO Tim Sabat.

Kicksaw’s technical resources love finding the perfect solution for our clients, but if you want to really spark joy, present them with a unique use case. Use case in point, a recent client ask: 

• As a user in Salesforce, I want to upload a resume and have it parsed for fields and objects I specify so that I don’t have to manually enter data

When we started looking into it, we discovered that a Salesforce native solution to complete this task doesn’t currently exist. Cue the excitement, our experts jumped into action. 

This blog will offer an overview of the custom AI resume parser we built, which reads documents for specific information in order to create records in Salesforce and will provide:

• A rundown of how it works
• Points of interest
• Examples of the other use cases this solution can easily support

Off-Platform AI + Salesforce: What We Built

To meet our client’s requirements and improve their solution’s ROI in a value-driven way we leveraged off-platform AI (in this case OpenAI), middleware on AWS (although we can use any cloud provider), flows, and the Salesforce platform’s flexible data architecture. 

Although our client’s use case involved resume parsing — it’s important to note that the way our solution functions is applicable to use cases across numerous industries, which we’ll cover in more detail later on in this blog.

How the solution works, using resume parsing as an example:

1. A resume is uploaded to the Salesforce platform (SFDC)
2. The resume is embedded using the middleware; the middleware is running as a web service  and works to add an additional layer of functionality and security
3. An AI prompt stored in the SFDC metadata is leveraged (Prompt example: From this document, create or update a Contact in Salesforce)
4. The middleware feeds the pdf resume to OpenAI as an embedding
5. The SDFC-supplied prompt  is enhanced by Python code to make it more specific to the task
6. The AI model parses the resume using the prompt as context, and returns only the necessary information
7. This result is consumed by the middleware and used to upsert a record in SFDC
8. The record, in this case a Contact, is ready for whatever downstream business logic is required of your application

From a user’s perspective this is done with a single click vs. manually scanning the document to identify the correct information, and then manually creating a record and entering the data needed. 

For the example of parsing resumes, you could use this solution to automatically create a record in Salesforce with details like:

• First and last name
• Job experience
• Language proficiencies
• Phone number and email address
• References
• Or any other field on your Contact record

Points of Interest and Additional Use Case Examples

Typically, in an integration, code for every field type in Salesforce has been mapped in advance and married to data from an api. The Kicksaw solution outlined above does away with this deterministic way of thinking. It tells the AI what fields we want filled in and lets it do the reasoning about how to map data from the resume to the fields.  

Salesforce has some innovative on-platform AI tools, but this off-platform AI solution’s functionality allows us to do some really creative things that extend Salesforce’s capabilities and fill in current gaps — making an organization’s SFDC that much more customizable. 

• Its contextual nature can be easily adapted to a wide variety of use cases 

• Record types aren’t confined, as the solution can just as easily create a Lead as it can a Contact, or any other record type

• The prompt and the SFDC metadata are the only things that change to adapt to various use cases 

• The prompt can easily be changed or updated by an Admin 

Possibilities for use cases are pretty endless, but a few examples include:

• For HLS: parsing intake forms to create patient records, using upgraded middleware for HIPAA compliance
  • Patient information can be taken from a patient-submitted document for information like medications to create a medication list in the contact record

• For Any Industry: parsing onboarding documentation for new hires, contractors, etc.
  • Parser can take employee forms with demographics, emergency contact, etc., and turn it into relevant information in SFDC

• For Service: parsing client intake PDFs to create contact records for SLAs
  • Easily transferring relevant data to records that assist with SLA enforcement

Still with us or are you as lost in thought as we are about what this solution and SFDC can do? Feel free to contact us if you have a use case in mind or would like to chat with us about this solution further.