Day 3 – Using GPT to Extract Key Fields from PDF Text #LaravelGPT #DocumentParsing #SmartContracts #AIExtraction #OpenAI #LaravelAI

Today we’ll send the extracted PDF text to GPT to pull out structured fields, such as parties involved, dates, amounts, and obligations — turning raw contracts/invoices into usable data.

Table of Contents

🎯 Goal

Parse AI-readable text into structured JSON using GPT function calling.

📦 Step 1: Define GPT Function Schema

Create a helper or inline array:

$functions = [
    [
        "name" => "extract_contract_details",
        "description" => "Extracts parties, dates, and values from contract or invoice text.",
        "parameters" => [
            "type" => "object",
            "properties" => [
                "party_a" => ["type" => "string"],
                "party_b" => ["type" => "string"],
                "start_date" => ["type" => "string"],
                "end_date" => ["type" => "string"],
                "total_amount" => ["type" => "string"],
            ],
            "required" => ["party_a", "party_b"]
        ],
    ]
];

🧠 Step 2: Send Extracted Text to GPT

In a new service DocumentAnalysisService.php:

use OpenAI\Laravel\Facades\OpenAI;

public function extractFields(string $text): array
{
    $response = OpenAI::chat()->create([
        'model' => 'gpt-4',
        'messages' => [
            ['role' => 'user', 'content' => "Extract structured fields from this text:\n\n" . $text],
        ],
        'functions' => $functions,
        'function_call' => ['name' => 'extract_contract_details'],
    ]);

    $result = $response['choices'][0]['message']['function_call']['arguments'] ?? '{}';

    return json_decode($result, true);
}

📥 Step 3: Update Controller to Save Extracted Data

Update DocumentController.php:

use App\Services\DocumentAnalysisService;

public function store(Request $request, PdfTextExtractor $extractor, DocumentAnalysisService $ai)
{
    ...
    $text = $extractor->extract($filename);
    $fields = $ai->extractFields($text);

    Document::create([
        'title' => $request->title,
        'type' => $request->type,
        'filename' => $filename,
        'user_id' => auth()->id(),
        'extracted_text' => $text,
        // optional: store $fields as JSON in another column or separate table
    ]);
}

Optional: create a document_fields table to store these fields if needed for searching/filtering.

✅ Summary

✅ Today you:

Defined a GPT function to extract key values
Sent raw PDF text to GPT for structured data extraction
Returned parsed JSON fields like parties and amounts

✅ Up next (Day 4): We’ll store and display extracted fields, and optionally allow admins to approve or correct GPT results before accepting them.

Post Views: 10