KI-gestützte PDF-Verarbeitung in C#: Summarize, Extrahieren und Analysieren von Dokumenten mit IronPDF

IronPdf.License.LicenseKey = "KEY";

IronPdf.License.LicenseKey = "KEY";

using IronPdf;
using IronPdf.AI;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Memory;
using Microsoft.SemanticKernel.Connectors.OpenAI;

// Initialize IronPDF AI with Azure OpenAI credentials

// Azure OpenAI configuration
string azureEndpoint = "https://your-resource.openai.azure.com/";
string apiKey = "your-azure-api-key";
string chatDeployment = "gpt-4o";
string embeddingDeployment = "text-embedding-ada-002";

// Initialize Semantic Kernel with Azure OpenAI
var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .AddAzureOpenAIChatCompletion(chatDeployment, azureEndpoint, apiKey)
    .Build();

// Create memory store for document embeddings
var memory = new MemoryBuilder()
    .WithMemoryStore(new VolatileMemoryStore())
    .WithAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .Build();

// Initialize IronPDF AI
IronDocumentAI.Initialize(kernel, memory);

Console.WriteLine("IronPDF AI initialized successfully with Azure OpenAI");

using IronPdf;
using IronPdf.AI;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Memory;
using Microsoft.SemanticKernel.Connectors.OpenAI;

// Initialize IronPDF AI with Azure OpenAI credentials

// Azure OpenAI configuration
string azureEndpoint = "https://your-resource.openai.azure.com/";
string apiKey = "your-azure-api-key";
string chatDeployment = "gpt-4o";
string embeddingDeployment = "text-embedding-ada-002";

// Initialize Semantic Kernel with Azure OpenAI
var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .AddAzureOpenAIChatCompletion(chatDeployment, azureEndpoint, apiKey)
    .Build();

// Create memory store for document embeddings
var memory = new MemoryBuilder()
    .WithMemoryStore(new VolatileMemoryStore())
    .WithAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .Build();

// Initialize IronPDF AI
IronDocumentAI.Initialize(kernel, memory);

Console.WriteLine("IronPDF AI initialized successfully with Azure OpenAI");

using IronPdf;
using IronPdf.AI;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Memory;
using Microsoft.SemanticKernel.Connectors.OpenAI;

// Summarize a PDF document using IronPDF AI

// Azure OpenAI configuration
string azureEndpoint = "https://your-resource.openai.azure.com/";
string apiKey = "your-azure-api-key";
string chatDeployment = "gpt-4o";
string embeddingDeployment = "text-embedding-ada-002";

// Initialize Semantic Kernel
var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .AddAzureOpenAIChatCompletion(chatDeployment, azureEndpoint, apiKey)
    .Build();

var memory = new MemoryBuilder()
    .WithMemoryStore(new VolatileMemoryStore())
    .WithAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .Build();

IronDocumentAI.Initialize(kernel, memory);

// Load and summarize PDF
var pdf = PdfDocument.FromFile("sample-report.pdf");
string summary = await pdf.Summarize();

Console.WriteLine("Document Summary:");
Console.WriteLine(summary);

File.WriteAllText("report-summary.txt", summary);
Console.WriteLine("\nSummary saved to report-summary.txt");

using IronPdf;
using IronPdf.AI;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Memory;
using Microsoft.SemanticKernel.Connectors.OpenAI;

// Summarize a PDF document using IronPDF AI

// Azure OpenAI configuration
string azureEndpoint = "https://your-resource.openai.azure.com/";
string apiKey = "your-azure-api-key";
string chatDeployment = "gpt-4o";
string embeddingDeployment = "text-embedding-ada-002";

// Initialize Semantic Kernel
var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .AddAzureOpenAIChatCompletion(chatDeployment, azureEndpoint, apiKey)
    .Build();

var memory = new MemoryBuilder()
    .WithMemoryStore(new VolatileMemoryStore())
    .WithAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .Build();

IronDocumentAI.Initialize(kernel, memory);

// Load and summarize PDF
var pdf = PdfDocument.FromFile("sample-report.pdf");
string summary = await pdf.Summarize();

Console.WriteLine("Document Summary:");
Console.WriteLine(summary);

File.WriteAllText("report-summary.txt", summary);
Console.WriteLine("\nSummary saved to report-summary.txt");

using IronPdf;
using IronPdf.AI;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Memory;
using Microsoft.SemanticKernel.Connectors.OpenAI;

// Synthesize insights across multiple related documents (e.g., quarterly reports into annual summary)

// Azure OpenAI configuration
string azureEndpoint = "https://your-resource.openai.azure.com/";
string apiKey = "your-azure-api-key";
string chatDeployment = "gpt-4o";
string embeddingDeployment = "text-embedding-ada-002";

// Initialize Semantic Kernel
var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .AddAzureOpenAIChatCompletion(chatDeployment, azureEndpoint, apiKey)
    .Build();

var memory = new MemoryBuilder()
    .WithMemoryStore(new VolatileMemoryStore())
    .WithAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .Build();

IronDocumentAI.Initialize(kernel, memory);

// Define documents to synthesize
string[] documentPaths = {
    "Q1-report.pdf",
    "Q2-report.pdf",
    "Q3-report.pdf",
    "Q4-report.pdf"
};

var documentSummaries = new List<string>();

// Summarize each document
foreach (string path in documentPaths)
{
    var pdf = PdfDocument.FromFile(path);
    string summary = await pdf.Summarize();
    documentSummaries.Add($"=== {Path.GetFileName(path)} ===\n{summary}");
    Console.WriteLine($"Processed: {path}");
}

// Combine and synthesize across all documents
string combinedSummaries = string.Join("\n\n", documentSummaries);

var synthesisDoc = PdfDocument.FromFile(documentPaths[0]);

string synthesisQuery = @"Based on the quarterly summaries below, provide an annual synthesis:
ll trends across quarters
chievements and challenges
over-year patterns

s:
inedSummaries;

string synthesis = await synthesisDoc.Query(synthesisQuery);

Console.WriteLine("\n=== Annual Synthesis ===");
Console.WriteLine(synthesis);

File.WriteAllText("annual-synthesis.txt", synthesis);

using IronPdf;
using IronPdf.AI;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Memory;
using Microsoft.SemanticKernel.Connectors.OpenAI;

// Synthesize insights across multiple related documents (e.g., quarterly reports into annual summary)

// Azure OpenAI configuration
string azureEndpoint = "https://your-resource.openai.azure.com/";
string apiKey = "your-azure-api-key";
string chatDeployment = "gpt-4o";
string embeddingDeployment = "text-embedding-ada-002";

// Initialize Semantic Kernel
var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .AddAzureOpenAIChatCompletion(chatDeployment, azureEndpoint, apiKey)
    .Build();

var memory = new MemoryBuilder()
    .WithMemoryStore(new VolatileMemoryStore())
    .WithAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .Build();

IronDocumentAI.Initialize(kernel, memory);

// Define documents to synthesize
string[] documentPaths = {
    "Q1-report.pdf",
    "Q2-report.pdf",
    "Q3-report.pdf",
    "Q4-report.pdf"
};

var documentSummaries = new List<string>();

// Summarize each document
foreach (string path in documentPaths)
{
    var pdf = PdfDocument.FromFile(path);
    string summary = await pdf.Summarize();
    documentSummaries.Add($"=== {Path.GetFileName(path)} ===\n{summary}");
    Console.WriteLine($"Processed: {path}");
}

// Combine and synthesize across all documents
string combinedSummaries = string.Join("\n\n", documentSummaries);

var synthesisDoc = PdfDocument.FromFile(documentPaths[0]);

string synthesisQuery = @"Based on the quarterly summaries below, provide an annual synthesis:
ll trends across quarters
chievements and challenges
over-year patterns

s:
inedSummaries;

string synthesis = await synthesisDoc.Query(synthesisQuery);

Console.WriteLine("\n=== Annual Synthesis ===");
Console.WriteLine(synthesis);

File.WriteAllText("annual-synthesis.txt", synthesis);

using IronPdf;
using IronPdf.AI;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Memory;
using Microsoft.SemanticKernel.Connectors.OpenAI;

// Generate executive summary from strategic documents for C-suite leadership

// Azure OpenAI configuration
string azureEndpoint = "https://your-resource.openai.azure.com/";
string apiKey = "your-azure-api-key";
string chatDeployment = "gpt-4o";
string embeddingDeployment = "text-embedding-ada-002";

// Initialize Semantic Kernel
var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .AddAzureOpenAIChatCompletion(chatDeployment, azureEndpoint, apiKey)
    .Build();

var memory = new MemoryBuilder()
    .WithMemoryStore(new VolatileMemoryStore())
    .WithAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .Build();

IronDocumentAI.Initialize(kernel, memory);

var pdf = PdfDocument.FromFile("strategic-plan.pdf");

string executiveQuery = @"Create an executive summary for C-suite leadership. Include:

cisions Required:**
ny decisions needing executive approval

al Findings:**
5 most important findings (bullet points)

ial Impact:**
e/cost implications if mentioned

ssessment:**
riority risks identified

ended Actions:**
ate next steps

er 500 words. Use business language appropriate for board presentation.";

string executiveSummary = await pdf.Query(executiveQuery);

File.WriteAllText("executive-summary.txt", executiveSummary);
Console.WriteLine("Executive summary saved to executive-summary.txt");

using IronPdf;
using IronPdf.AI;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Memory;
using Microsoft.SemanticKernel.Connectors.OpenAI;

// Generate executive summary from strategic documents for C-suite leadership

// Azure OpenAI configuration
string azureEndpoint = "https://your-resource.openai.azure.com/";
string apiKey = "your-azure-api-key";
string chatDeployment = "gpt-4o";
string embeddingDeployment = "text-embedding-ada-002";

// Initialize Semantic Kernel
var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .AddAzureOpenAIChatCompletion(chatDeployment, azureEndpoint, apiKey)
    .Build();

var memory = new MemoryBuilder()
    .WithMemoryStore(new VolatileMemoryStore())
    .WithAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .Build();

IronDocumentAI.Initialize(kernel, memory);

var pdf = PdfDocument.FromFile("strategic-plan.pdf");

string executiveQuery = @"Create an executive summary for C-suite leadership. Include:

cisions Required:**
ny decisions needing executive approval

al Findings:**
5 most important findings (bullet points)

ial Impact:**
e/cost implications if mentioned

ssessment:**
riority risks identified

ended Actions:**
ate next steps

er 500 words. Use business language appropriate for board presentation.";

string executiveSummary = await pdf.Query(executiveQuery);

File.WriteAllText("executive-summary.txt", executiveSummary);
Console.WriteLine("Executive summary saved to executive-summary.txt");

using IronPdf;
using IronPdf.AI;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Memory;
using Microsoft.SemanticKernel.Connectors.OpenAI;
using System.Text.Json;

// Extract structured invoice data as JSON from PDF

// Azure OpenAI configuration
string azureEndpoint = "https://your-resource.openai.azure.com/";
string apiKey = "your-azure-api-key";
string chatDeployment = "gpt-4o";
string embeddingDeployment = "text-embedding-ada-002";

// Initialize Semantic Kernel
var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .AddAzureOpenAIChatCompletion(chatDeployment, azureEndpoint, apiKey)
    .Build();

var memory = new MemoryBuilder()
    .WithMemoryStore(new VolatileMemoryStore())
    .WithAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .Build();

IronDocumentAI.Initialize(kernel, memory);

var pdf = PdfDocument.FromFile("sample-invoice.pdf");

// Define JSON schema for extraction
string extractionQuery = @"Extract invoice data and return as JSON with this exact structure:

voiceNumber"": ""string"",
voiceDate"": ""YYYY-MM-DD"",
eDate"": ""YYYY-MM-DD"",
ndor"": {
""name"": ""string"",
""address"": ""string"",
""taxId"": ""string or null""

stomer"": {
""name"": ""string"",
""address"": ""string""

neItems"": [
{
    ""description"": ""string"",
    ""quantity"": number,
    ""unitPrice"": number,
    ""total"": number
}

btotal"": number,
xRate"": number,
xAmount"": number,
tal"": number,
rrency"": ""string""


NLY valid JSON, no additional text.";

string jsonResponse = await pdf.Query(extractionQuery);

// Parse and save JSON
try
{
    var invoiceData = JsonSerializer.Deserialize<JsonElement>(jsonResponse);
    string formattedJson = JsonSerializer.Serialize(invoiceData, new JsonSerializerOptions { WriteIndented = true });

    Console.WriteLine("Extracted Invoice Data:");
    Console.WriteLine(formattedJson);

    File.WriteAllText("invoice-data.json", formattedJson);
}
catch (JsonException)
{
    Console.WriteLine("Unable to parse JSON response");
    File.WriteAllText("invoice-raw-response.txt", jsonResponse);
}

using IronPdf;
using IronPdf.AI;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Memory;
using Microsoft.SemanticKernel.Connectors.OpenAI;
using System.Text.Json;

// Extract structured invoice data as JSON from PDF

// Azure OpenAI configuration
string azureEndpoint = "https://your-resource.openai.azure.com/";
string apiKey = "your-azure-api-key";
string chatDeployment = "gpt-4o";
string embeddingDeployment = "text-embedding-ada-002";

// Initialize Semantic Kernel
var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .AddAzureOpenAIChatCompletion(chatDeployment, azureEndpoint, apiKey)
    .Build();

var memory = new MemoryBuilder()
    .WithMemoryStore(new VolatileMemoryStore())
    .WithAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .Build();

IronDocumentAI.Initialize(kernel, memory);

var pdf = PdfDocument.FromFile("sample-invoice.pdf");

// Define JSON schema for extraction
string extractionQuery = @"Extract invoice data and return as JSON with this exact structure:

voiceNumber"": ""string"",
voiceDate"": ""YYYY-MM-DD"",
eDate"": ""YYYY-MM-DD"",
ndor"": {
""name"": ""string"",
""address"": ""string"",
""taxId"": ""string or null""

stomer"": {
""name"": ""string"",
""address"": ""string""

neItems"": [
{
    ""description"": ""string"",
    ""quantity"": number,
    ""unitPrice"": number,
    ""total"": number
}

btotal"": number,
xRate"": number,
xAmount"": number,
tal"": number,
rrency"": ""string""


NLY valid JSON, no additional text.";

string jsonResponse = await pdf.Query(extractionQuery);

// Parse and save JSON
try
{
    var invoiceData = JsonSerializer.Deserialize<JsonElement>(jsonResponse);
    string formattedJson = JsonSerializer.Serialize(invoiceData, new JsonSerializerOptions { WriteIndented = true });

    Console.WriteLine("Extracted Invoice Data:");
    Console.WriteLine(formattedJson);

    File.WriteAllText("invoice-data.json", formattedJson);
}
catch (JsonException)
{
    Console.WriteLine("Unable to parse JSON response");
    File.WriteAllText("invoice-raw-response.txt", jsonResponse);
}

using IronPdf;
using IronPdf.AI;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Memory;
using Microsoft.SemanticKernel.Connectors.OpenAI;
using System.Text.Json;

// Analyze contract clauses and identify key terms, risks, and critical dates

// Azure OpenAI configuration
string azureEndpoint = "https://your-resource.openai.azure.com/";
string apiKey = "your-azure-api-key";
string chatDeployment = "gpt-4o";
string embeddingDeployment = "text-embedding-ada-002";

// Initialize Semantic Kernel
var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .AddAzureOpenAIChatCompletion(chatDeployment, azureEndpoint, apiKey)
    .Build();

var memory = new MemoryBuilder()
    .WithMemoryStore(new VolatileMemoryStore())
    .WithAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .Build();

IronDocumentAI.Initialize(kernel, memory);

var pdf = PdfDocument.FromFile("contract.pdf");

// Define JSON schema for contract analysis
string clauseQuery = @"Analyze this contract and identify key clauses. Return JSON:

ntractType"": ""string"",
rties"": [""string""],
fectiveDate"": ""string"",
auses"": [
{
    ""type"": ""Termination|Liability|Indemnification|Confidentiality|IP|Payment|Warranty|Other"",
    ""title"": ""string"",
    ""summary"": ""string"",
    ""riskLevel"": ""Low|Medium|High"",
    ""keyTerms"": [""string""]
}

iticalDates"": [
{
    ""description"": ""string"",
    ""date"": ""string""
}

erallRiskAssessment"": ""Low|Medium|High"",
commendations"": [""string""]


: termination rights, liability caps, indemnification, IP ownership, confidentiality, payment terms.
NLY valid JSON.";

string analysisJson = await pdf.Query(clauseQuery);

try
{
    var analysis = JsonSerializer.Deserialize<JsonElement>(analysisJson);
    string formatted = JsonSerializer.Serialize(analysis, new JsonSerializerOptions { WriteIndented = true });

    Console.WriteLine("Contract Clause Analysis:");
    Console.WriteLine(formatted);

    File.WriteAllText("contract-analysis.json", formatted);

    // Display high-risk clauses
    Console.WriteLine("\n=== High Risk Clauses ===");
    foreach (var clause in analysis.GetProperty("clauses").EnumerateArray())
    {
        if (clause.GetProperty("riskLevel").GetString() == "High")
        {
            Console.WriteLine($"- {clause.GetProperty("type")}: {clause.GetProperty("summary")}");
        }
    }
}
catch (JsonException)
{
    Console.WriteLine("Unable to parse contract analysis");
    File.WriteAllText("contract-analysis-raw.txt", analysisJson);
}

using IronPdf;
using IronPdf.AI;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Memory;
using Microsoft.SemanticKernel.Connectors.OpenAI;
using System.Text.Json;

// Analyze contract clauses and identify key terms, risks, and critical dates

// Azure OpenAI configuration
string azureEndpoint = "https://your-resource.openai.azure.com/";
string apiKey = "your-azure-api-key";
string chatDeployment = "gpt-4o";
string embeddingDeployment = "text-embedding-ada-002";

// Initialize Semantic Kernel
var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .AddAzureOpenAIChatCompletion(chatDeployment, azureEndpoint, apiKey)
    .Build();

var memory = new MemoryBuilder()
    .WithMemoryStore(new VolatileMemoryStore())
    .WithAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .Build();

IronDocumentAI.Initialize(kernel, memory);

var pdf = PdfDocument.FromFile("contract.pdf");

// Define JSON schema for contract analysis
string clauseQuery = @"Analyze this contract and identify key clauses. Return JSON:

ntractType"": ""string"",
rties"": [""string""],
fectiveDate"": ""string"",
auses"": [
{
    ""type"": ""Termination|Liability|Indemnification|Confidentiality|IP|Payment|Warranty|Other"",
    ""title"": ""string"",
    ""summary"": ""string"",
    ""riskLevel"": ""Low|Medium|High"",
    ""keyTerms"": [""string""]
}

iticalDates"": [
{
    ""description"": ""string"",
    ""date"": ""string""
}

erallRiskAssessment"": ""Low|Medium|High"",
commendations"": [""string""]


: termination rights, liability caps, indemnification, IP ownership, confidentiality, payment terms.
NLY valid JSON.";

string analysisJson = await pdf.Query(clauseQuery);

try
{
    var analysis = JsonSerializer.Deserialize<JsonElement>(analysisJson);
    string formatted = JsonSerializer.Serialize(analysis, new JsonSerializerOptions { WriteIndented = true });

    Console.WriteLine("Contract Clause Analysis:");
    Console.WriteLine(formatted);

    File.WriteAllText("contract-analysis.json", formatted);

    // Display high-risk clauses
    Console.WriteLine("\n=== High Risk Clauses ===");
    foreach (var clause in analysis.GetProperty("clauses").EnumerateArray())
    {
        if (clause.GetProperty("riskLevel").GetString() == "High")
        {
            Console.WriteLine($"- {clause.GetProperty("type")}: {clause.GetProperty("summary")}");
        }
    }
}
catch (JsonException)
{
    Console.WriteLine("Unable to parse contract analysis");
    File.WriteAllText("contract-analysis-raw.txt", analysisJson);
}

using IronPdf;
using IronPdf.AI;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Memory;
using Microsoft.SemanticKernel.Connectors.OpenAI;
using System.Text.Json;

// Extract financial metrics from annual reports and earnings documents

// Azure OpenAI configuration
string azureEndpoint = "https://your-resource.openai.azure.com/";
string apiKey = "your-azure-api-key";
string chatDeployment = "gpt-4o";
string embeddingDeployment = "text-embedding-ada-002";

// Initialize Semantic Kernel
var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .AddAzureOpenAIChatCompletion(chatDeployment, azureEndpoint, apiKey)
    .Build();

var memory = new MemoryBuilder()
    .WithMemoryStore(new VolatileMemoryStore())
    .WithAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .Build();

IronDocumentAI.Initialize(kernel, memory);

var pdf = PdfDocument.FromFile("annual-report.pdf");

// Define JSON schema for financial extraction (numbers in millions)
string financialQuery = @"Extract financial metrics from this document. Return JSON:

portPeriod"": ""string"",
mpany"": ""string"",
rrency"": ""string"",
comeStatement"": {
""revenue"": number,
""costOfRevenue"": number,
""grossProfit"": number,
""operatingExpenses"": number,
""operatingIncome"": number,
""netIncome"": number,
""eps"": number

lanceSheet"": {
""totalAssets"": number,
""totalLiabilities"": number,
""shareholdersEquity"": number,
""cash"": number,
""totalDebt"": number

yMetrics"": {
""revenueGrowthYoY"": ""string"",
""grossMargin"": ""string"",
""operatingMargin"": ""string"",
""netMargin"": ""string"",
""debtToEquity"": number

idance"": {
""nextQuarterRevenue"": ""string"",
""fullYearRevenue"": ""string"",
""notes"": ""string""



 for unavailable data. Numbers in millions unless stated.
NLY valid JSON.";

string financialJson = await pdf.Query(financialQuery);

try
{
    var financials = JsonSerializer.Deserialize<JsonElement>(financialJson);
    string formatted = JsonSerializer.Serialize(financials, new JsonSerializerOptions { WriteIndented = true });

    Console.WriteLine("Extracted Financial Data:");
    Console.WriteLine(formatted);

    File.WriteAllText("financial-data.json", formatted);
}
catch (JsonException)
{
    Console.WriteLine("Unable to parse financial data");
    File.WriteAllText("financial-raw.txt", financialJson);
}

using IronPdf;
using IronPdf.AI;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Memory;
using Microsoft.SemanticKernel.Connectors.OpenAI;
using System.Text.Json;

// Extract financial metrics from annual reports and earnings documents

// Azure OpenAI configuration
string azureEndpoint = "https://your-resource.openai.azure.com/";
string apiKey = "your-azure-api-key";
string chatDeployment = "gpt-4o";
string embeddingDeployment = "text-embedding-ada-002";

// Initialize Semantic Kernel
var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .AddAzureOpenAIChatCompletion(chatDeployment, azureEndpoint, apiKey)
    .Build();

var memory = new MemoryBuilder()
    .WithMemoryStore(new VolatileMemoryStore())
    .WithAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .Build();

IronDocumentAI.Initialize(kernel, memory);

var pdf = PdfDocument.FromFile("annual-report.pdf");

// Define JSON schema for financial extraction (numbers in millions)
string financialQuery = @"Extract financial metrics from this document. Return JSON:

portPeriod"": ""string"",
mpany"": ""string"",
rrency"": ""string"",
comeStatement"": {
""revenue"": number,
""costOfRevenue"": number,
""grossProfit"": number,
""operatingExpenses"": number,
""operatingIncome"": number,
""netIncome"": number,
""eps"": number

lanceSheet"": {
""totalAssets"": number,
""totalLiabilities"": number,
""shareholdersEquity"": number,
""cash"": number,
""totalDebt"": number

yMetrics"": {
""revenueGrowthYoY"": ""string"",
""grossMargin"": ""string"",
""operatingMargin"": ""string"",
""netMargin"": ""string"",
""debtToEquity"": number

idance"": {
""nextQuarterRevenue"": ""string"",
""fullYearRevenue"": ""string"",
""notes"": ""string""



 for unavailable data. Numbers in millions unless stated.
NLY valid JSON.";

string financialJson = await pdf.Query(financialQuery);

try
{
    var financials = JsonSerializer.Deserialize<JsonElement>(financialJson);
    string formatted = JsonSerializer.Serialize(financials, new JsonSerializerOptions { WriteIndented = true });

    Console.WriteLine("Extracted Financial Data:");
    Console.WriteLine(formatted);

    File.WriteAllText("financial-data.json", formatted);
}
catch (JsonException)
{
    Console.WriteLine("Unable to parse financial data");
    File.WriteAllText("financial-raw.txt", financialJson);
}

using IronPdf;
using IronPdf.AI;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Memory;
using Microsoft.SemanticKernel.Connectors.OpenAI;
using System.Text.Json;

// Extract structured research metadata from academic papers

// Azure OpenAI configuration
string azureEndpoint = "https://your-resource.openai.azure.com/";
string apiKey = "your-azure-api-key";
string chatDeployment = "gpt-4o";
string embeddingDeployment = "text-embedding-ada-002";

// Initialize Semantic Kernel
var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .AddAzureOpenAIChatCompletion(chatDeployment, azureEndpoint, apiKey)
    .Build();

var memory = new MemoryBuilder()
    .WithMemoryStore(new VolatileMemoryStore())
    .WithAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .Build();

IronDocumentAI.Initialize(kernel, memory);

var pdf = PdfDocument.FromFile("research-paper.pdf");

// Define JSON schema for research paper extraction
string researchQuery = @"Extract structured information from this research paper. Return JSON:

tle"": ""string"",
thors"": [""string""],
stitution"": ""string"",
blicationDate"": ""string"",
stract"": ""string"",
searchQuestion"": ""string"",
thodology"": {
""type"": ""Quantitative|Qualitative|Mixed Methods"",
""approach"": ""string"",
""sampleSize"": ""string"",
""dataCollection"": ""string""

yFindings"": [
{
    ""finding"": ""string"",
    ""significance"": ""string"",
    ""confidence"": ""High|Medium|Low""
}

mitations"": [""string""],
tureWork"": [""string""],
ywords"": [""string""]


 extracting verifiable claims and noting uncertainty.
NLY valid JSON.";

string extractionResult = await pdf.Query(researchQuery);

try
{
    var research = JsonSerializer.Deserialize<JsonElement>(extractionResult);
    string formatted = JsonSerializer.Serialize(research, new JsonSerializerOptions { WriteIndented = true });

    Console.WriteLine("Research Paper Extraction:");
    Console.WriteLine(formatted);

    File.WriteAllText("research-extraction.json", formatted);

    // Display key findings with confidence levels
    Console.WriteLine("\n=== Key Findings ===");
    foreach (var finding in research.GetProperty("keyFindings").EnumerateArray())
    {
        string confidence = finding.GetProperty("confidence").GetString() ?? "Unknown";
        Console.WriteLine($"[{confidence}] {finding.GetProperty("finding")}");
    }
}
catch (JsonException)
{
    Console.WriteLine("Unable to parse research extraction");
    File.WriteAllText("research-raw.txt", extractionResult);
}

using IronPdf;
using IronPdf.AI;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Memory;
using Microsoft.SemanticKernel.Connectors.OpenAI;
using System.Text.Json;

// Extract structured research metadata from academic papers

// Azure OpenAI configuration
string azureEndpoint = "https://your-resource.openai.azure.com/";
string apiKey = "your-azure-api-key";
string chatDeployment = "gpt-4o";
string embeddingDeployment = "text-embedding-ada-002";

// Initialize Semantic Kernel
var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .AddAzureOpenAIChatCompletion(chatDeployment, azureEndpoint, apiKey)
    .Build();

var memory = new MemoryBuilder()
    .WithMemoryStore(new VolatileMemoryStore())
    .WithAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .Build();

IronDocumentAI.Initialize(kernel, memory);

var pdf = PdfDocument.FromFile("research-paper.pdf");

// Define JSON schema for research paper extraction
string researchQuery = @"Extract structured information from this research paper. Return JSON:

tle"": ""string"",
thors"": [""string""],
stitution"": ""string"",
blicationDate"": ""string"",
stract"": ""string"",
searchQuestion"": ""string"",
thodology"": {
""type"": ""Quantitative|Qualitative|Mixed Methods"",
""approach"": ""string"",
""sampleSize"": ""string"",
""dataCollection"": ""string""

yFindings"": [
{
    ""finding"": ""string"",
    ""significance"": ""string"",
    ""confidence"": ""High|Medium|Low""
}

mitations"": [""string""],
tureWork"": [""string""],
ywords"": [""string""]


 extracting verifiable claims and noting uncertainty.
NLY valid JSON.";

string extractionResult = await pdf.Query(researchQuery);

try
{
    var research = JsonSerializer.Deserialize<JsonElement>(extractionResult);
    string formatted = JsonSerializer.Serialize(research, new JsonSerializerOptions { WriteIndented = true });

    Console.WriteLine("Research Paper Extraction:");
    Console.WriteLine(formatted);

    File.WriteAllText("research-extraction.json", formatted);

    // Display key findings with confidence levels
    Console.WriteLine("\n=== Key Findings ===");
    foreach (var finding in research.GetProperty("keyFindings").EnumerateArray())
    {
        string confidence = finding.GetProperty("confidence").GetString() ?? "Unknown";
        Console.WriteLine($"[{confidence}] {finding.GetProperty("finding")}");
    }
}
catch (JsonException)
{
    Console.WriteLine("Unable to parse research extraction");
    File.WriteAllText("research-raw.txt", extractionResult);
}

using IronPdf;
using IronPdf.AI;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Memory;
using Microsoft.SemanticKernel.Connectors.OpenAI;

// Interactive Q&A system for querying PDF documents

// Azure OpenAI configuration
string azureEndpoint = "https://your-resource.openai.azure.com/";
string apiKey = "your-azure-api-key";
string chatDeployment = "gpt-4o";
string embeddingDeployment = "text-embedding-ada-002";

// Initialize Semantic Kernel
var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .AddAzureOpenAIChatCompletion(chatDeployment, azureEndpoint, apiKey)
    .Build();

var memory = new MemoryBuilder()
    .WithMemoryStore(new VolatileMemoryStore())
    .WithAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .Build();

IronDocumentAI.Initialize(kernel, memory);

var pdf = PdfDocument.FromFile("sample-legal-document.pdf");

// Memorize document to enable persistent querying
await pdf.Memorize();

Console.WriteLine("PDF Q&A System - Type 'exit' to quit\n");
Console.WriteLine($"Document loaded and memorized: {pdf.PageCount} pages\n");

// Interactive Q&A loop
while (true)
{
    Console.Write("Your question: ");
    string? question = Console.ReadLine();

    if (string.IsNullOrWhiteSpace(question) || question.ToLower() == "exit")
        break;

    string answer = await pdf.Query(question);

    Console.WriteLine($"\nAnswer: {answer}\n");
    Console.WriteLine(new string('-', 50) + "\n");
}

Console.WriteLine("Q&A session ended.");

using IronPdf;
using IronPdf.AI;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Memory;
using Microsoft.SemanticKernel.Connectors.OpenAI;

// Interactive Q&A system for querying PDF documents

// Azure OpenAI configuration
string azureEndpoint = "https://your-resource.openai.azure.com/";
string apiKey = "your-azure-api-key";
string chatDeployment = "gpt-4o";
string embeddingDeployment = "text-embedding-ada-002";

// Initialize Semantic Kernel
var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .AddAzureOpenAIChatCompletion(chatDeployment, azureEndpoint, apiKey)
    .Build();

var memory = new MemoryBuilder()
    .WithMemoryStore(new VolatileMemoryStore())
    .WithAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .Build();

IronDocumentAI.Initialize(kernel, memory);

var pdf = PdfDocument.FromFile("sample-legal-document.pdf");

// Memorize document to enable persistent querying
await pdf.Memorize();

Console.WriteLine("PDF Q&A System - Type 'exit' to quit\n");
Console.WriteLine($"Document loaded and memorized: {pdf.PageCount} pages\n");

// Interactive Q&A loop
while (true)
{
    Console.Write("Your question: ");
    string? question = Console.ReadLine();

    if (string.IsNullOrWhiteSpace(question) || question.ToLower() == "exit")
        break;

    string answer = await pdf.Query(question);

    Console.WriteLine($"\nAnswer: {answer}\n");
    Console.WriteLine(new string('-', 50) + "\n");
}

Console.WriteLine("Q&A session ended.");

using IronPdf;

// Split long documents into overlapping chunks for RAG systems

var pdf = PdfDocument.FromFile("long-document.pdf");

// Chunking configuration
int maxChunkTokens = 4000;      // Leave room for prompts and responses
int overlapTokens = 200;        // Overlap for context continuity
int approxCharsPerToken = 4;    // Rough estimate for tokenization

int maxChunkChars = maxChunkTokens * approxCharsPerToken;
int overlapChars = overlapTokens * approxCharsPerToken;

var chunks = new List<DocumentChunk>();
var currentChunk = new System.Text.StringBuilder();
int chunkStartPage = 1;
int currentPage = 1;

for (int i = 0; i < pdf.PageCount; i++)
{
    string pageText = pdf.Pages[i].Text;
    currentPage = i + 1;

    if (currentChunk.Length + pageText.Length > maxChunkChars && currentChunk.Length > 0)
    {
        chunks.Add(new DocumentChunk
        {
            Text = currentChunk.ToString(),
            StartPage = chunkStartPage,
            EndPage = currentPage - 1,
            ChunkIndex = chunks.Count
        });

        // Create overlap with previous chunk for continuity
        string overlap = currentChunk.Length > overlapChars
            ? currentChunk.ToString().Substring(currentChunk.Length - overlapChars)
            : currentChunk.ToString();

        currentChunk.Clear();
        currentChunk.Append(overlap);
        chunkStartPage = currentPage - 1;
    }

    currentChunk.AppendLine($"\n--- Page {currentPage} ---\n");
    currentChunk.Append(pageText);
}

if (currentChunk.Length > 0)
{
    chunks.Add(new DocumentChunk
    {
        Text = currentChunk.ToString(),
        StartPage = chunkStartPage,
        EndPage = currentPage,
        ChunkIndex = chunks.Count
    });
}

Console.WriteLine($"Document chunked into {chunks.Count} segments");
foreach (var chunk in chunks)
{
    Console.WriteLine($"  Chunk {chunk.ChunkIndex + 1}: Pages {chunk.StartPage}-{chunk.EndPage} ({chunk.Text.Length} chars)");
}

// Save chunk metadata for RAG indexing
File.WriteAllText("chunks-metadata.json", System.Text.Json.JsonSerializer.Serialize(
    chunks.Select(c => new { c.ChunkIndex, c.StartPage, c.EndPage, Length = c.Text.Length }),
    new System.Text.Json.JsonSerializerOptions { WriteIndented = true }
));


ic class DocumentChunk

public string Text { get; set; } = "";
public int StartPage { get; set; }
public int EndPage { get; set; }
public int ChunkIndex { get; set; }

using IronPdf;

// Split long documents into overlapping chunks for RAG systems

var pdf = PdfDocument.FromFile("long-document.pdf");

// Chunking configuration
int maxChunkTokens = 4000;      // Leave room for prompts and responses
int overlapTokens = 200;        // Overlap for context continuity
int approxCharsPerToken = 4;    // Rough estimate for tokenization

int maxChunkChars = maxChunkTokens * approxCharsPerToken;
int overlapChars = overlapTokens * approxCharsPerToken;

var chunks = new List<DocumentChunk>();
var currentChunk = new System.Text.StringBuilder();
int chunkStartPage = 1;
int currentPage = 1;

for (int i = 0; i < pdf.PageCount; i++)
{
    string pageText = pdf.Pages[i].Text;
    currentPage = i + 1;

    if (currentChunk.Length + pageText.Length > maxChunkChars && currentChunk.Length > 0)
    {
        chunks.Add(new DocumentChunk
        {
            Text = currentChunk.ToString(),
            StartPage = chunkStartPage,
            EndPage = currentPage - 1,
            ChunkIndex = chunks.Count
        });

        // Create overlap with previous chunk for continuity
        string overlap = currentChunk.Length > overlapChars
            ? currentChunk.ToString().Substring(currentChunk.Length - overlapChars)
            : currentChunk.ToString();

        currentChunk.Clear();
        currentChunk.Append(overlap);
        chunkStartPage = currentPage - 1;
    }

    currentChunk.AppendLine($"\n--- Page {currentPage} ---\n");
    currentChunk.Append(pageText);
}

if (currentChunk.Length > 0)
{
    chunks.Add(new DocumentChunk
    {
        Text = currentChunk.ToString(),
        StartPage = chunkStartPage,
        EndPage = currentPage,
        ChunkIndex = chunks.Count
    });
}

Console.WriteLine($"Document chunked into {chunks.Count} segments");
foreach (var chunk in chunks)
{
    Console.WriteLine($"  Chunk {chunk.ChunkIndex + 1}: Pages {chunk.StartPage}-{chunk.EndPage} ({chunk.Text.Length} chars)");
}

// Save chunk metadata for RAG indexing
File.WriteAllText("chunks-metadata.json", System.Text.Json.JsonSerializer.Serialize(
    chunks.Select(c => new { c.ChunkIndex, c.StartPage, c.EndPage, Length = c.Text.Length }),
    new System.Text.Json.JsonSerializerOptions { WriteIndented = true }
));


ic class DocumentChunk

public string Text { get; set; } = "";
public int StartPage { get; set; }
public int EndPage { get; set; }
public int ChunkIndex { get; set; }

using IronPdf;
using IronPdf.AI;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Memory;
using Microsoft.SemanticKernel.Connectors.OpenAI;

// Retrieval-Augmented Generation (RAG) system for querying across multiple indexed documents

// Azure OpenAI configuration
string azureEndpoint = "https://your-resource.openai.azure.com/";
string apiKey = "your-azure-api-key";
string chatDeployment = "gpt-4o";
string embeddingDeployment = "text-embedding-ada-002";

// Initialize Semantic Kernel
var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .AddAzureOpenAIChatCompletion(chatDeployment, azureEndpoint, apiKey)
    .Build();

var memory = new MemoryBuilder()
    .WithMemoryStore(new VolatileMemoryStore())
    .WithAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .Build();

IronDocumentAI.Initialize(kernel, memory);

// Index all documents in folder
string[] documentPaths = Directory.GetFiles("documents/", "*.pdf");

Console.WriteLine($"Indexing {documentPaths.Length} documents...\n");

// Memorize each document (creates embeddings for retrieval)
foreach (string path in documentPaths)
{
    var pdf = PdfDocument.FromFile(path);
    await pdf.Memorize();
    Console.WriteLine($"Indexed: {Path.GetFileName(path)} ({pdf.PageCount} pages)");
}

Console.WriteLine("\n=== RAG System Ready ===\n");

// Query across all indexed documents
string query = "What are the key compliance requirements for data retention?";

Console.WriteLine($"Query: {query}\n");

var searchPdf = PdfDocument.FromFile(documentPaths[0]);
string answer = await searchPdf.Query(query);

Console.WriteLine($"Answer: {answer}");

// Interactive query loop
Console.WriteLine("\n--- Enter questions (type 'exit' to quit) ---\n");

while (true)
{
    Console.Write("Question: ");
    string? userQuery = Console.ReadLine();

    if (string.IsNullOrWhiteSpace(userQuery) || userQuery.ToLower() == "exit")
        break;

    string response = await searchPdf.Query(userQuery);
    Console.WriteLine($"\nAnswer: {response}\n");
}

using IronPdf;
using IronPdf.AI;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Memory;
using Microsoft.SemanticKernel.Connectors.OpenAI;

// Retrieval-Augmented Generation (RAG) system for querying across multiple indexed documents

// Azure OpenAI configuration
string azureEndpoint = "https://your-resource.openai.azure.com/";
string apiKey = "your-azure-api-key";
string chatDeployment = "gpt-4o";
string embeddingDeployment = "text-embedding-ada-002";

// Initialize Semantic Kernel
var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .AddAzureOpenAIChatCompletion(chatDeployment, azureEndpoint, apiKey)
    .Build();

var memory = new MemoryBuilder()
    .WithMemoryStore(new VolatileMemoryStore())
    .WithAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .Build();

IronDocumentAI.Initialize(kernel, memory);

// Index all documents in folder
string[] documentPaths = Directory.GetFiles("documents/", "*.pdf");

Console.WriteLine($"Indexing {documentPaths.Length} documents...\n");

// Memorize each document (creates embeddings for retrieval)
foreach (string path in documentPaths)
{
    var pdf = PdfDocument.FromFile(path);
    await pdf.Memorize();
    Console.WriteLine($"Indexed: {Path.GetFileName(path)} ({pdf.PageCount} pages)");
}

Console.WriteLine("\n=== RAG System Ready ===\n");

// Query across all indexed documents
string query = "What are the key compliance requirements for data retention?";

Console.WriteLine($"Query: {query}\n");

var searchPdf = PdfDocument.FromFile(documentPaths[0]);
string answer = await searchPdf.Query(query);

Console.WriteLine($"Answer: {answer}");

// Interactive query loop
Console.WriteLine("\n--- Enter questions (type 'exit' to quit) ---\n");

while (true)
{
    Console.Write("Question: ");
    string? userQuery = Console.ReadLine();

    if (string.IsNullOrWhiteSpace(userQuery) || userQuery.ToLower() == "exit")
        break;

    string response = await searchPdf.Query(userQuery);
    Console.WriteLine($"\nAnswer: {response}\n");
}

using IronPdf;
using IronPdf.AI;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Memory;
using Microsoft.SemanticKernel.Connectors.OpenAI;
using System.Text.RegularExpressions;

// Answer questions with page citations and source verification

// Azure OpenAI configuration
string azureEndpoint = "https://your-resource.openai.azure.com/";
string apiKey = "your-azure-api-key";
string chatDeployment = "gpt-4o";
string embeddingDeployment = "text-embedding-ada-002";

// Initialize Semantic Kernel
var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .AddAzureOpenAIChatCompletion(chatDeployment, azureEndpoint, apiKey)
    .Build();

var memory = new MemoryBuilder()
    .WithMemoryStore(new VolatileMemoryStore())
    .WithAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .Build();

IronDocumentAI.Initialize(kernel, memory);

var pdf = PdfDocument.FromFile("sample-legal-document.pdf");
await pdf.Memorize();

string question = "What are the termination conditions in this agreement?";

// Request citations in query
string citationQuery = $@"{question}

T: Include specific page citations in your answer using the format (Page X) or (Pages X-Y).
e information that appears in the document.";

string answerWithCitations = await pdf.Query(citationQuery);

Console.WriteLine("Question: " + question);
Console.WriteLine("\nAnswer with Citations:");
Console.WriteLine(answerWithCitations);

// Extract cited page numbers using regex
var citedPages = ExtractCitedPages(answerWithCitations);
Console.WriteLine($"\nCited pages: {string.Join(", ", citedPages)}");

// Verify citations with page excerpts
Console.WriteLine("\n=== Source Verification ===");
foreach (int pageNum in citedPages.Take(3))
{
    if (pageNum <= pdf.PageCount && pageNum > 0)
    {
        string pageText = pdf.Pages[pageNum - 1].Text;
        string excerpt = pageText.Length > 200 ? pageText.Substring(0, 200) + "..." : pageText;
        Console.WriteLine($"\nPage {pageNum} excerpt:\n{excerpt}");
    }
}

// Extract page numbers from citation format (Page X) or (Pages X-Y)
List<int> ExtractCitedPages(string text)
{
    var pages = new HashSet<int>();
    var matches = Regex.Matches(text, @"\(Pages?\s*(\d+)(?:\s*-\s*(\d+))?\)", RegexOptions.IgnoreCase);

    foreach (Match match in matches)
    {
        int startPage = int.Parse(match.Groups[1].Value);
        pages.Add(startPage);

        if (match.Groups[2].Success)
        {
            int endPage = int.Parse(match.Groups[2].Value);
            for (int p = startPage; p <= endPage; p++)
                pages.Add(p);
        }
    }
    return pages.OrderBy(p => p).ToList();
}

using IronPdf;
using IronPdf.AI;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Memory;
using Microsoft.SemanticKernel.Connectors.OpenAI;
using System.Text.RegularExpressions;

// Answer questions with page citations and source verification

// Azure OpenAI configuration
string azureEndpoint = "https://your-resource.openai.azure.com/";
string apiKey = "your-azure-api-key";
string chatDeployment = "gpt-4o";
string embeddingDeployment = "text-embedding-ada-002";

// Initialize Semantic Kernel
var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .AddAzureOpenAIChatCompletion(chatDeployment, azureEndpoint, apiKey)
    .Build();

var memory = new MemoryBuilder()
    .WithMemoryStore(new VolatileMemoryStore())
    .WithAzureOpenAITextEmbeddingGeneration(embeddingDeployment, azureEndpoint, apiKey)
    .Build();

IronDocumentAI.Initialize(kernel, memory);

var pdf = PdfDocument.FromFile("sample-legal-document.pdf");
await pdf.Memorize();

string question = "What are the termination conditions in this agreement?";

// Request citations in query
string citationQuery = $@"{question}

T: Include specific page citations in your answer using the format (Page X) or (Pages X-Y).
e information that appears in the document.";

string answerWithCitations = await pdf.Query(citationQuery);

Console.WriteLine("Question: " + question);
Console.WriteLine("\nAnswer with Citations:");
Console.WriteLine(answerWithCitations);

// Extract cited page numbers using regex
var citedPages = ExtractCitedPages(answerWithCitations);
Console.WriteLine($"\nCited pages: {string.Join(", ", citedPages)}");

// Verify citations with page excerpts
Console.WriteLine("\n=== Source Verification ===");
foreach (int pageNum in citedPages.Take(3))
{
    if (pageNum <= pdf.PageCount && pageNum > 0)
    {
        string pageText = pdf.Pages[pageNum - 1].Text;
        string excerpt = pageText.Length > 200 ? pageText.Substring(0, 200) + "..." : pageText;
        Console.WriteLine($"\nPage {pageNum} excerpt:\n{excerpt}");
    }
}

// Extract page numbers from citation format (Page X) or (Pages X-Y)
List<int> ExtractCitedPages(string text)
{
    var pages = new HashSet<int>();
    var matches = Regex.Matches(text, @"\(Pages?\s*(\d+)(?:\s*-\s*(\d+))?\)", RegexOptions.IgnoreCase);

    foreach (Match match in matches)
    {
        int startPage = int.Parse(match.Groups[1].Value);
        pages.Add(startPage);

        if (match.Groups[2].Success)
        {
            int endPage = int.Parse(match.Groups[2].Value);
            for (int p = startPage; p <= endPage; p++)
                pages.Add(p);
        }
    }
    return pages.OrderBy(p => p).ToList();
}