Skip to content

Latest commit

 

History

History
376 lines (278 loc) · 11.5 KB

README.md

File metadata and controls

376 lines (278 loc) · 11.5 KB
Raw

Need more flexibility? Try the Extractor package instead, a AI-Powered data extraction library for Laravel

AI-Powered Receipt and Invoice Scanner for Laravel

Latest Version on Packagist Total Downloads

Easily extract structured receipt data from images, PDFs, and emails within your Laravel application using OpenAI.

Features

  • Light wrapper around OpenAI Chat and Completion endpoints.
  • Accepts text as input and returns structured receipt information.
  • Includes a well-tuned prompt for parsing receipts.
  • Supports various input formats including Plain Text, PDF, Images, Word documents, and Web content.
  • Integrates with Textract for OCR functionality.

Installation

Install the package via composer:

composer require helgesverre/receipt-scanner

Publish the config file:

php artisan vendor:publish --tag="receipt-scanner-config"

All the configuration options are documented in the configuration file.

Since this package uses the OpenAI Laravel Package, so you also need to publish their config and add the OPENAI_API_KEY to your .env file:

php artisan vendor:publish --provider="OpenAI\Laravel\ServiceProvider"
OPENAI_API_KEY="your-key-here

Usage

Extracting receipt data from Plain Text

Plain text scanning is useful when you already have the textual representation of a receipt or invoice.

The example is from a Paddle.com receipt email, where I copied all the text in the email, and removed all the empty lines.

$text = <<<RECEIPT
Liseth Solutions AS
via software reseller Paddle.com
Thank you for your purchase!
Your full invoice is attached to this email.
Amount paid
Payment method
NOK 2,498.75
visa
ending in 4242
Test: SaaS Subscription - Pro Plan
September 22, 2023 11:04 am UTC - October 22, 2023 11:04 am UTC
NOK 1,999.00
QTY: 1
Subtotal
NOK 1,999.00
VAT
NOK 499.75
Amount paid*
NOK 2,498.75
*This payment will appear on your statement as: PADDLE.NET* EXAMPLEINC
NEED HELP?
Need help with your purchase? Please contact us on paddle.net.
logo
Paddle.com Market Ltd, Judd House, 18-29 Mora Street, London EC1V 8BT
© 2023 Paddle. All rights reserved.
RECEIPT;


ReceiptScanner::scan($text);

Extracting data from other formats

use HelgeSverre\ReceiptScanner\Facades\Text;

$textPlainText = Text::text(file_get_contents('./receipt.txt'));
$textPdf = Text::pdf(file_get_contents('./receipt.pdf'));
$textImageOcr = Text::textract(file_get_contents('./receipt.jpg'));
$textPdfOcr = Text::textractUsingS3Upload(file_get_contents('./receipt.pdf'));
$textWord = Text::word(file_get_contents('./receipt.doc'));
$textWeb = Text::web('https://example.com');
$textHtml = Text::html(file_get_contents('./receipt.html'));

After loading, you can pass the TextContent or the plain text (which can be retrieved by calling ->toString()) into the ReceiptScanner::scan() method.

use HelgeSverre\ReceiptScanner\Facades\ReceiptScanner;

ReceiptScanner::scan($textPlainText)
ReceiptScanner::scan($textPdf)
ReceiptScanner::scan($textImageOcr)
ReceiptScanner::scan($textPdfOcr)
ReceiptScanner::scan($textWord)
ReceiptScanner::scan($textWeb)
ReceiptScanner::scan($textHtml)

Receipt Data Model

The scanned receipt is parsed into a DTO which consists of a main Receipt class, which contains the receipt metadata, and a Merchant dto, representing the seller on the receipt or invoice, and an array of LineItem DTOs holding each individual line item.

  • HelgeSverre\ReceiptScanner\Data\Receipt
  • HelgeSverre\ReceiptScanner\Data\Merchant
  • HelgeSverre\ReceiptScanner\Data\LineItem

The DTO has a toArray() method, which will result in a structure like this:

For flexibility, all fields are nullable.

[
    "orderRef" => "string",
    "date" => "date",
    "taxAmount" => "number",
    "totalAmount" => "number",
    "currency" => "string",
    "merchant" => [
        "name" => "string",
        "vatId" => "string",
        "address" => "string",
    ],
    "lineItems" => [
        [
            "text" => "string",
            "sku" => "string",
            "qty" => "number",
            "price" => "number",
        ],
    ],
];

Returning an Array instead of a DTO

If you prefer to work with an array instead of the built-in DTO, you can specify asArray: true when calling scan()

use HelgeSverre\ReceiptScanner\Facades\ReceiptScanner;

ReceiptScanner::scan(
    $textPlainText
    asArray: true
)

Specifying the model

To use a different model, you can specify the model name to use with the model named argument when calling the scan() method.

use HelgeSverre\ReceiptScanner\Facades\ReceiptScanner;
use HelgeSverre\ReceiptScanner\ModelNames;

// With the ModelNames class
ReceiptScanner::scan($content, model: ModelNames::GPT4_1106_PREVIEW)

// With a string
ReceiptScanner::scan($content, model: 'gpt-4-1106-preview')

All parameters and what they do

$text (TextContent|string)

The input text from the receipt or invoice that needs to be parsed. It accepts either a TextContent object or a string.

**$model (string)

This parameter specifies the OpenAI model used for the extraction process.

HelgeSverre\ReceiptScanner\ModelNames is a class containing constants for each model, provided for convenience. However, you can also directly use a string to specify the model if you prefer.

Different models have different speed/accuracy characteristics.

If you require high accuracy, use a GPT-4 model, if you need speed, use a GPT-3 model, if you need even more speed, use the gpt-3.5-turbo-instruct model.

The default model is ModelNames::TURBO_INSTRUCT.

ModelNames Constant Value
ModelNames::TURBO gpt-3.5-turbo
ModelNames::TURBO_INSTRUCT gpt-3.5-turbo-instruct
ModelNames::TURBO_1106 gpt-3.5-turbo-1106
ModelNames::TURBO_16K gpt-3.5-turbo-16k
ModelNames::TURBO_0613 gpt-3.5-turbo-0613
ModelNames::TURBO_16K_0613 gpt-3.5-turbo-16k-0613
ModelNames::TURBO_0301 gpt-3.5-turbo-0301
ModelNames::GPT4 gpt-4
ModelNames::GPT4_32K gpt-4-32k
ModelNames::GPT4_32K_0613 gpt-4-32k-0613
ModelNames::GPT4_1106_PREVIEW gpt-4-1106-preview
ModelNames::GPT4_0314 gpt-4-0314
ModelNames::GPT4_32K_0314 gpt-4-32k-0314

$maxTokens (int)

The maximum number of tokens that the model will processes. The default value is 2000, adjusting this value may be necessary for very long text, but 2000 is "usually" fairly good.

$temperature (float)

Controls the randomness/creativity of the model's output.

A higher value (e.g., 0.8) makes the output more random, which is usually not what we want in this scenario, I usually go with 0.1 or 0.2, anything over 0.5 becomes useless. Defaults to 0.1.

$template (string)

This parameter specifies the template used for the prompt.

The default template is 'receipt'. You can create and use additional templates by adding new blade files in the resources/views/vendor/receipt-scanner/ directory and specifying the file name (without extension) as the $template value (eg: "minimal_invoice".

$asArray (bool)

If true, returns the response from the AI model as an array instead of as a DTO, useful if you need to modifythe default DTO to have more/less fields or want to convert the response into your own DTO, defaults to false

Example Usage:

use HelgeSverre\ReceiptScanner\Facades\ReceiptScanner;

$parsedReceipt = ReceiptScanner::scan(
    text: $textInput,
    model: ModelNames::TURBO_INSTRUCT,
    maxTokens: 500,
    temperature: 0.2,
    template: 'minimal_invoice',
    asArray: true,
);

List of supported models

Enum Value Model name Endpoint
TURBO_INSTRUCT gpt-3.5-turbo-instruct Completion
TURBO_16K gpt-3.5-turbo-16k Chat
TURBO gpt-3.5-turbo Chat
GPT4 gpt-4 Chat
GPT4_32K gpt-4-32 Chat

OCR Configuration with AWS Textract

To use AWS Textract for extracting text from large images and multi-page PDFs, the package needs to upload the file to S3 and pass the s3 object location along to the textract service.

So you need to configure your AWS Credentials in the config/receipt-scanner.php file as follows:

TEXTRACT_KEY="your-aws-access-key"
TEXTRACT_SECRET="your-aws-security"
TEXTRACT_REGION="your-textract-region"

# Can be omitted
TEXTRACT_VERSION="2018-06-27"

You also need to configure a seperate Textract disk where the files will be stored, open your config/filesystems.php configuration file and add the following:

'textract' => [
    'driver' => 's3',
    'key' => env('TEXTRACT_KEY'),
    'secret' => env('TEXTRACT_SECRET'),
    'region' => env('TEXTRACT_REGION'),
    'bucket' => env('TEXTRACT_BUCKET'),
],

Ensure the textract_disk setting in config/receipt-scanner.php is the same as your disk name in the filesystems.php config, you can change it with the .env value TEXTRACT_DISK.

return [
    "textract_disk" => env("TEXTRACT_DISK")
];

.env

TEXTRACT_DISK="uploads"

Note

Textract is not available in all regions:

Q: In which AWS regions is Amazon Textract available? Amazon Textract is currently available in the US East (Northern Virginia), US East (Ohio), US West (Oregon), US West ( N. California), AWS GovCloud (US-West), AWS GovCloud (US-East), Canada (Central), EU (Ireland), EU (London), EU ( Frankfurt), EU (Paris), Asia Pacific (Singapore), Asia Pacific (Sydney), Asia Pacific (Seoul), and Asia Pacific ( Mumbai) Regions.

See: https://aws.amazon.com/textract/faqs/

Publishing Prompts

You may publish the prompt file that is used under the hood by running this command:

php artisan vendor:publish --tag="receipt-scanner-prompts"

This package simply uses blade files as prompts, the {{ $context }} variable will be replaced by the text you pass to ReceiptScanner::scan("text here").

Adding prompts/templates

By default, the package uses the receipt.blade.php file as its prompt template, you may add additional templates by simply creating a blade file in resources/views/vendor/receipt-scanner/minimal_invoice.blade.php and changing the $template parameter when calling scan()

Example prompt:

Extract the following fields from the text below, output as JSON

date (as string in the  Y-m-d format)
total_amount (as float, do not include currency symbol) 
vendor_name (company name)

{{ $context }}

OUTPUT IN JSON
use HelgeSverre\ReceiptScanner\Facades\ReceiptScanner;

$receipt = ReceiptScanner::scan(
    text: "Your invoice here",
    model:  ModelNames::TURBO_INSTRUCT,
    template: 'minimal_invoice',
    asArray: true,
);

License

This package is licensed under the MIT License. For more details, refer to the License File.