GPT enrichment setup

This page describes how you can setup a GPT enrichment in a few steps using the Metamaze endpoint.

Follow these easy steps to set up an enrichment to use OpenAI's GPT.

Prerequisites: OpenAI credentials

You will need the following OpenAI credentials:

API Key
Organisation ID

You can find and generate these in the OpenAI platform:

1. Go to the enrichments project settings and press the create button

2. Fill in the general tab

Metadata

For the metadata there are 2 different configurations: one on the entity level and the other document level.

Type: Entity

{
  "enrichment_name": "Force Majeure",
  "enrichment_type": "entity",
  "entity_name": [
    "Force Majeure Clause"
  ],
  "prompt": "Is it possible that the agreement is terminated, due to effects of Force Majeure?",
  "answers": [
    "Yes",
    "No"
  ],
  "OPENAI_API_KEY": "xxxxxxxx",
  "OPENAI_API_ORG": "xxxxxxxx"
}

Field Description

Field	Description
enrichment_name	has to be exactly the same as the name of the enrichment in project settings
enrichment_type	fixed value: `'entity'`
entity_names	names of the entity to enrich
prompt	Prompt for GPT. The task to send as instruction to GPT model.
answers	[Optional] It’s an optional field (you can skip it) to define the exact answers with which GPT can answer. However, we would recommend defining short, clear answers like "Yes", "No" for 3 reasons. Mainly, it allows for a bigger chance that GPT will respond correctly with the exact desired answer and also ensures easier automatic processing. It is also cheaper since fewer tokens are used.
OPENAI_API_KEY	API Key (see )
OPENAI_API_ORG	Organisation ID (see )

enrichment_name

has to be exactly the same as the name of the enrichment in project settings

enrichment_type

fixed value: 'entity'

entity_names

names of the entity to enrich

prompt

Prompt for GPT. The task to send as instruction to GPT model.

answers

[Optional] It’s an optional field (you can skip it) to define the exact answers with which GPT can answer. However, we would recommend defining short, clear answers like "Yes", "No" for 3 reasons. Mainly, it allows for a bigger chance that GPT will respond correctly with the exact desired answer and also ensures easier automatic processing. It is also cheaper since fewer tokens are used.

OPENAI_API_KEY

API Key (see )

OPENAI_API_ORG

Organisation ID (see )

Type: Document

{
  "enrichment_name": "Currency",
  "enrichment_type": "document",
  "prompt": "In which currency is the document?",
  "answers": [
    "€",
    "$"
  ],
  "OPENAI_API_KEY": "xxxxxxxx",
  "OPENAI_API_ORG": "xxxxxxxx"
}