Custom document AI analysis

Introduction

Some documents can only be judged against what you already know: does this power of attorney name the right signatory, for the right company? Custom document AI analysis lets you write those conditions as a prompt, and checks each uploaded document against the case's own data. It can approve or reject on the result, or leave the decision to you.

How to use it

Configuring a Prompt

Define the validation criteria in a prompt.
The prompt should specify the key elements to be checked, such as name consistency, date accuracy, or other relevant data points.

Setting Auto-Approval or Auto-Rejection

You can enable automatic approval if all criteria are met.
Alternatively, automatic rejection can be activated if any criteria fail.
If auto-approval or auto-rejection is disabled, manual reviews will be required on the document check.

Testing the Prompt

Test the prompt by uploading a sample document.
Select an entity in the case to serve as the test reference data.
The validation process checks if the extracted data meets the conditions specified in the prompt, in comparison with the reference data of the selected entity within the case.
Review the validation results to fine-tune the prompt as needed.

Example prompts

Example of custom document	Example of prompt
Power of attorney	- Validate that this is a valid power of attorney. Giving the power from this context individual to the context signatory to open an account for the context company. Containing a signature or mention related to a signature.
Business Agreement with Liquidity Partners	- Verify that this document is an agreement with a liquidity provider.
Corporate / Group Structure Chart	- Verify that this document is a corporate / group structure chart, that it is not older than 6 months and that it has been signed by a director of the entity
Proof of Business Evidence	- Verify that this document is an example of agreements with customers and/or partners showcasing the scope of services

🤔
Most frequently asked questions

Who are your LLM providers?
➡️ Dotfile partners with OpenAI and Mistral AI. Customers can choose their preferred provider.

Where is/are the data/documents stored depending on the provider?
➡️ Documents are exclusively stored in AWS/s3 (EU) regardless of the chosen LLM provider.

Are the files uploaded used for training the model?
➡️ Documents are OCR'd by Amazon Textract, which does not store them. The extracted text goes to the LLM provider over the API, which retains nothing. Nothing you upload is used to train a model.