README

Nette balik pro vytezovani faktur a ucetnich dokladu z PDF (vcetne skenu) pres:

• Azure Document Intelligence (OCR + strukturovana extrakce)
• LLM normalizaci (Azure OpenAI)
• ceske regex fallbacky (VS, DUZP, ICO, DIC)
• validacni vrstvu a asynchronni worker pattern

Instalace

composer require insightbase/invoice-parser-nette

Azure setup (API key + endpoint + deployment)

Niz je doporuceny postup. Urceno pro stav k 15. 3. 2026.

1) Azure ucet a subscription

1. Vytvor nebo pouzij existujici Azure account.
2. Over, ze mas aktivni subscription a opravneni aspon Contributor na resource group.

2) Azure Document Intelligence (azureDi)

1. V Azure Portal vytvor resource typu Document Intelligence (historicky Form Recognizer).
2. Vyber region, kde sluzbu chces provozovat.
3. Po vytvoreni otevri Keys and Endpoint.
4. Zkopiruj:

• Endpoint -> pouzij jako AZURE_DI_ENDPOINT
• Key 1 nebo Key 2 -> pouzij jako AZURE_DI_KEY

3) Azure OpenAI (llm)

1. V Azure Portal vytvor resource Azure OpenAI.
2. V resource otevri Keys and Endpoint.
3. Zkopiruj:

• Endpoint -> AZURE_OPENAI_ENDPOINT
• Key 1 nebo Key 2 -> AZURE_OPENAI_KEY

1. Otevri Azure AI Foundry / model deployment panel pro tento resource.
2. Vytvor model deployment (napr. GPT model) a zapamatuj deployment name -> AZURE_OPENAI_DEPLOYMENT.

Poznamka:

• Pokud nejde Azure OpenAI resource nebo deployment vytvorit, jde obvykle o chybejici quota/permission v tenantu nebo regionu. V tom pripade je potreba pozadat Azure admina o povoleni.

4) Promenne prostredi

Minimalne nastav:

AZURE_DI_ENDPOINT=https://<your-di-resource>.cognitiveservices.azure.com
AZURE_DI_KEY=<your-di-key>
AZURE_OPENAI_ENDPOINT=https://<your-openai-resource>.openai.azure.com
AZURE_OPENAI_KEY=<your-openai-key>
AZURE_OPENAI_DEPLOYMENT=<your-deployment-name>

5) Konfigurace extension v Nette

extensions:
    invoiceParser: InsightBase\InvoiceParserNette\DI\InvoiceParserExtension

invoiceParser:
    azureDi:
        endpoint: %env(AZURE_DI_ENDPOINT)%
        apiKey: %env(AZURE_DI_KEY)%
        model: prebuilt-invoice
        apiVersion: 2023-07-31
        maxPollAttempts: 25
        pollIntervalMs: 1000
    llm:
        enabled: true
        endpoint: %env(AZURE_OPENAI_ENDPOINT)%
        deployment: %env(AZURE_OPENAI_DEPLOYMENT)%
        apiKey: %env(AZURE_OPENAI_KEY)%
        apiVersion: 2024-10-21

6) Odkazy na oficialni dokumentaci

• Azure Document Intelligence quickstart: https://learn.microsoft.com/azure/ai-services/document-intelligence/quickstarts/get-started-sdks-rest-api
• Azure OpenAI chat completions quickstart: https://learn.microsoft.com/azure/ai-services/openai/chatgpt-quickstart
• Azure OpenAI role-based access control: https://learn.microsoft.com/azure/ai-services/openai/how-to/role-based-access-control

Pouziti

<?php

declare(strict_types=1);

use InsightBase\InvoiceParserNette\Parser\InvoiceParser;

final class InvoiceService
{
    public function __construct(
        private InvoiceParser $invoiceParser,
    ) {
    }

    public function parse(string $pdfPath): array
    {
        $pdfContent = file_get_contents($pdfPath);
        $result = $this->invoiceParser->parsePdf((string) $pdfContent);

        return $result->invoice->toArray();
    }
}

Asynchronni worker (Contributte RabbitMQ)

Knihovna obsahuje worker service InvoiceParseWorker::process(array $message).

Priklad payloadu zpravy:

{
  "pdfPath": "/data/invoices/invoice-2026-001.pdf"
}

Nebo:

{
  "pdfBase64": "JVBERi0xLjQKJ..."
}

Ukazkova integrace je v examples/rabbitmq.neon a examples/InvoiceConsumer.php.

Poznamky

• Pro oskenovane PDF se OCR resi na strane Azure Document Intelligence.
• Regex fallback slouzi jako doplnek, kdyz DI/LLM vrati neuplna data.
• Validator hlida zakladni konzistenci castek a dat.