GPT Step Getting Started

🌄 Flow

1. Start of with the prompt template. The prompt template should include:
   1. The signals you want to extract
   2. The data-context the model should use to extract data. (This will be done using functions)
2. After you have the prompt fill in signals.
   1. For each signal you need to fill in:
      1. Name - Will be the signal name
      2. Description - Explain the model exactly how to compute this step.
      3. Type - Boolean / String.
3. Run

🐍 Functions

In order to add data context to the prompt we use functions. Functions are executed and added to the prompt prior to sending the prompt to the model.

Supported functions

Search

Parameters: query
Description: Performs a Google search using the query string parameter and returns the results.

Here is an example where we might wand to add context regarding company's layoff:

Here is some relevant context:  
Company website:  
{{Search(query=record['Webiste'] + ' layoffs')}}

ReadUrlText

ReadUrlText(link): Retrieves and returns the text content from a webpage given its link.

## Example
Assuming dataset as a column named domain

Company search results:  
{{ReadUrlText(link=record['domain'])}}

LinkedinSearch

SearchLinkedin(query): Searches for specific entities (such as profiles or companies) on 
LinkedIn based on the query.

Linkedin search results:
{{SearchLinkedin(query=record['domain'] + ' ceo')}}

GetCompanyContacts

GetCompanyContacts(domain, job_titles, skills): Extracts and returns the contact information
of a company specified by domain, filtered by job_titles and skills.

LinkQuestionAnswering

LinkQuestionAnswering(link, question): Analyzes a question, maps it to
a data model within the webpage at the provided url, and retrieves the answer.

SearchForNaicsCodes

SearchForNaicsCodes(query, naics_code_level: Optional): 
Searches for NAICS codes based on provided query (e.g., “cloud security”, “clothing ecommerce”).