AI-Powered PDF Processing in C#：使用 IronPDF 总结、提取和分析文档

PM > Install-Package IronPdf
PM > Install-Package IronPdf.Extensions.AI
PM > Install-Package Microsoft.SemanticKernel
PM > Install-Package Microsoft.SemanticKernel.Plugins.Memory

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;

// 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;
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 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;

// 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;
using System;
using System.Collections.Concurrent;
using System.Text;

// Process multiple documents in parallel with rate limiting

// 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);

// Configure parallel processing with rate limiting
int maxConcurrency = 3;
string inputFolder = "documents/";
string outputFolder = "summaries/";

Directory.CreateDirectory(outputFolder);

string[] pdfFiles = Directory.GetFiles(inputFolder, "*.pdf");
Console.WriteLine($"Processing {pdfFiles.Length} documents...\n");

var results = new ConcurrentBag<ProcessingResult>();
var semaphore = new SemaphoreSlim(maxConcurrency);

var tasks = pdfFiles.Select(async filePath =>
{
    await semaphore.WaitAsync();
    var result = new ProcessingResult { FilePath = filePath };

    try
    {
        var stopwatch = System.Diagnostics.Stopwatch.StartNew();

        var pdf = PdfDocument.FromFile(filePath);
        string summary = await pdf.Summarize();

        string outputPath = Path.Combine(outputFolder,
            Path.GetFileNameWithoutExtension(filePath) + "-summary.txt");
        await File.WriteAllTextAsync(outputPath, summary);

        stopwatch.Stop();
        result.Success = true;
        result.ProcessingTime = stopwatch.Elapsed;
        result.OutputPath = outputPath;

        Console.WriteLine($"[OK] {Path.GetFileName(filePath)} ({stopwatch.ElapsedMilliseconds}ms)");
    }
    catch (Exception ex)
    {
        result.Success = false;
        result.ErrorMessage = ex.Message;
        Console.WriteLine($"[ERROR] {Path.GetFileName(filePath)}: {ex.Message}");
    }
    finally
    {
        semaphore.Release();
        results.Add(result);
    }
}).ToArray();

await Task.WhenAll(tasks);

// Generate processing report
var successful = results.Where(r => r.Success).ToList();
var failed = results.Where(r => !r.Success).ToList();

var report = new StringBuilder();
report.AppendLine("=== Batch Processing Report ===");
report.AppendLine($"Successful: {successful.Count}");
report.AppendLine($"Failed: {failed.Count}");

if (successful.Any())
{
    var avgTime = TimeSpan.FromMilliseconds(successful.Average(r => r.ProcessingTime.TotalMilliseconds));
    report.AppendLine($"Average processing time: {avgTime.TotalSeconds:F1}s");
}

if (failed.Any())
{
    report.AppendLine("\nFailed documents:");
    foreach (var fail in failed)
        report.AppendLine($"  - {Path.GetFileName(fail.FilePath)}: {fail.ErrorMessage}");
}

string reportText = report.ToString();
Console.WriteLine($"\n{reportText}");
File.WriteAllText(Path.Combine(outputFolder, "processing-report.txt"), reportText);


s ProcessingResult

public string FilePath { get; set; } = "";
public bool Success { get; set; }
public TimeSpan ProcessingTime { get; set; }
public string OutputPath { get; set; } = "";
public string ErrorMessage { get; set; } = "";

using IronPdf;
using System.Text.Json;
using System.Net.Http.Headers;

// Use OpenAI Batch API for 50% cost savings on large-scale document processing

string openAiApiKey = "your-openai-api-key";
string inputFolder = "documents/";

// Prepare batch requests in JSONL format
var batchRequests = new List<string>();
string[] pdfFiles = Directory.GetFiles(inputFolder, "*.pdf");

Console.WriteLine($"Preparing batch for {pdfFiles.Length} documents...\n");

foreach (string filePath in pdfFiles)
{
    var pdf = PdfDocument.FromFile(filePath);
    string pdfText = pdf.ExtractAllText();

    // Truncate to stay within batch API limits
    if (pdfText.Length > 100000)
        pdfText = pdfText.Substring(0, 100000) + "\n[Truncated...]";

    var request = new
    {
        custom_id = Path.GetFileNameWithoutExtension(filePath),
        method = "POST",
        url = "/v1/chat/completions",
        body = new
        {
            model = "gpt-4o",
            messages = new[]
            {
                new { role = "system", content = "Summarize the following document concisely." },
                new { role = "user", content = pdfText }
            },
            max_tokens = 1000
        }
    };

    batchRequests.Add(JsonSerializer.Serialize(request));
}

// Create JSONL file
string batchFilePath = "batch-requests.jsonl";
File.WriteAllLines(batchFilePath, batchRequests);
Console.WriteLine($"Created batch file with {batchRequests.Count} requests");

// Upload file to OpenAI
using var httpClient = new HttpClient();
httpClient.DefaultRequestHeaders.Authorization = new AuthenticationHeaderValue("Bearer", openAiApiKey);

using var fileContent = new MultipartFormDataContent();
fileContent.Add(new ByteArrayContent(File.ReadAllBytes(batchFilePath)), "file", "batch-requests.jsonl");
fileContent.Add(new StringContent("batch"), "purpose");

var uploadResponse = await httpClient.PostAsync("https://api.openai.com/v1/files", fileContent);
var uploadResult = JsonSerializer.Deserialize<JsonElement>(await uploadResponse.Content.ReadAsStringAsync());
string fileId = uploadResult.GetProperty("id").GetString()!;
Console.WriteLine($"Uploaded file: {fileId}");

// Create batch job (24-hour completion window for 50% discount)
var batchJobRequest = new
{
    input_file_id = fileId,
    endpoint = "/v1/chat/completions",
    completion_window = "24h"
};

var batchResponse = await httpClient.PostAsync(
    "https://api.openai.com/v1/batches",
    new StringContent(JsonSerializer.Serialize(batchJobRequest), System.Text.Encoding.UTF8, "application/json")
);

var batchResult = JsonSerializer.Deserialize<JsonElement>(await batchResponse.Content.ReadAsStringAsync());
string batchId = batchResult.GetProperty("id").GetString()!;

Console.WriteLine($"\nBatch job created: {batchId}");
Console.WriteLine("Job will complete within 24 hours");
Console.WriteLine($"Check status: GET https://api.openai.com/v1/batches/{batchId}");

File.WriteAllText("batch-job-id.txt", batchId);
Console.WriteLine("\nBatch ID saved to batch-job-id.txt");

using IronPdf;
using IronPdf.AI;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Memory;
using Microsoft.SemanticKernel.Connectors.OpenAI;
using System;
using System.Collections.Generic;
using System.Security.Cryptography;
using System.Text.Json;

// Cache AI processing results using file hashes to avoid reprocessing unchanged 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);

// Configure caching
string cacheFolder = "ai-cache/";
string documentsFolder = "documents/";

Directory.CreateDirectory(cacheFolder);

var cacheManager = new DocumentCacheManager(cacheFolder);

// Process documents with caching
string[] pdfFiles = Directory.GetFiles(documentsFolder, "*.pdf");
int cached = 0, processed = 0;

foreach (string filePath in pdfFiles)
{
    string fileName = Path.GetFileName(filePath);
    string fileHash = cacheManager.ComputeFileHash(filePath);

    var cachedResult = cacheManager.GetCachedResult(fileName, fileHash);

    if (cachedResult != null)
    {
        Console.WriteLine($"[CACHE HIT] {fileName}");
        cached++;
        continue;
    }

    Console.WriteLine($"[PROCESSING] {fileName}");
    var pdf = PdfDocument.FromFile(filePath);
    string summary = await pdf.Summarize();

    cacheManager.CacheResult(fileName, fileHash, summary);
    processed++;
}

Console.WriteLine($"\nProcessing complete: {cached} cached, {processed} newly processed");
Console.WriteLine($"Cost savings: {(cached * 100.0 / Math.Max(1, cached + processed)):F1}% served from cache");


ash-based cache manager with JSON index
s DocumentCacheManager

private readonly string _cacheFolder;
private readonly string _indexPath;
private Dictionary<string, CacheEntry> _index;

public DocumentCacheManager(string cacheFolder)
{
    _cacheFolder = cacheFolder;
    _indexPath = Path.Combine(cacheFolder, "cache-index.json");
    _index = LoadIndex();
}

private Dictionary<string, CacheEntry> LoadIndex()
{
    if (File.Exists(_indexPath))
    {
        string json = File.ReadAllText(_indexPath);
        return JsonSerializer.Deserialize<Dictionary<string, CacheEntry>>(json) ?? new();
    }
    return new Dictionary<string, CacheEntry>();
}

private void SaveIndex()
{
    string json = JsonSerializer.Serialize(_index, new JsonSerializerOptions { WriteIndented = true });
    File.WriteAllText(_indexPath, json);
}

// SHA256 hash to detect file changes
public string ComputeFileHash(string filePath)
{
    using var sha256 = SHA256.Create();
    using var stream = File.OpenRead(filePath);
    byte[] hash = sha256.ComputeHash(stream);
    return Convert.ToHexString(hash);
}

public string? GetCachedResult(string fileName, string currentHash)
{
    if (_index.TryGetValue(fileName, out var entry))
    {
        if (entry.FileHash == currentHash && File.Exists(entry.CachePath))
        {
            entry.LastAccessed = DateTime.UtcNow;
            SaveIndex();
            return File.ReadAllText(entry.CachePath);
        }
    }
    return null;
}

public void CacheResult(string fileName, string fileHash, string result)
{
    string cachePath = Path.Combine(_cacheFolder, $"{Path.GetFileNameWithoutExtension(fileName)}-{fileHash[..8]}.txt");
    File.WriteAllText(cachePath, result);

    _index[fileName] = new CacheEntry
    {
        FileHash = fileHash,
        CachePath = cachePath,
        CreatedAt = DateTime.UtcNow,
        LastAccessed = DateTime.UtcNow
    };

    SaveIndex();
}


s CacheEntry

public string FileHash { get; set; } = "";
public string CachePath { get; set; } = "";
public DateTime CreatedAt { get; set; }
public DateTime LastAccessed { get; set; }

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

// Compare financial metrics across multiple company filings for sector analysis

// 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);

// Analyze company filings
string[] companyFilings = {
    "filings/company-a-10k.pdf",
    "filings/company-b-10k.pdf",
    "filings/company-c-10k.pdf"
};

var sectorData = new List<CompanyFinancials>();

foreach (string filing in companyFilings)
{
    Console.WriteLine($"Analyzing: {Path.GetFileName(filing)}");

    var pdf = PdfDocument.FromFile(filing);

    // Define JSON schema for 10-K extraction (numbers in millions USD)
    string extractionQuery = @"Extract key financial metrics from this 10-K filing. Return JSON:

mpanyName"": ""string"",
scalYear"": ""string"",
venue"": number,
venueGrowth"": number,
ossMargin"": number,
eratingMargin"": number,
tIncome"": number,
s"": number,
talDebt"": number,
shPosition"": number,
ployeeCount"": number,
yRisks"": [""string""],
idance"": ""string""


in millions USD. Growth/margins as percentages.
NLY valid JSON.";

    string result = await pdf.Query(extractionQuery);

    try
    {
        var financials = JsonSerializer.Deserialize<CompanyFinancials>(result);
        if (financials != null)
            sectorData.Add(financials);
    }
    catch
    {
        Console.WriteLine($"  Warning: Could not parse financials for {filing}");
    }
}

// Generate sector comparison report
var report = new StringBuilder();
report.AppendLine("=== Sector Analysis Report ===\n");

report.AppendLine("Revenue Comparison (millions USD):");
foreach (var company in sectorData.OrderByDescending(c => c.Revenue))
    report.AppendLine($"  {company.CompanyName}: ${company.Revenue:N0} ({company.RevenueGrowth:+0.0;-0.0}% YoY)");

report.AppendLine("\nProfitability Margins:");
foreach (var company in sectorData.OrderByDescending(c => c.OperatingMargin))
    report.AppendLine($"  {company.CompanyName}: {company.GrossMargin:F1}% gross, {company.OperatingMargin:F1}% operating");

report.AppendLine("\nFinancial Health (Debt vs Cash):");
foreach (var company in sectorData)
{
    double netDebt = company.TotalDebt - company.CashPosition;
    string status = netDebt < 0 ? "Net Cash" : "Net Debt";
    report.AppendLine($"  {company.CompanyName}: {status} ${Math.Abs(netDebt):N0}M");
}

string reportText = report.ToString();
Console.WriteLine($"\n{reportText}");
File.WriteAllText("sector-analysis-report.txt", reportText);

// Save full JSON data
string outputJson = JsonSerializer.Serialize(sectorData, new JsonSerializerOptions { WriteIndented = true });
File.WriteAllText("sector-analysis.json", outputJson);

Console.WriteLine("Analysis saved to sector-analysis.json and sector-analysis-report.txt");


s CompanyFinancials

public string CompanyName { get; set; } = "";
public string FiscalYear { get; set; } = "";
public double Revenue { get; set; }
public double RevenueGrowth { get; set; }
public double GrossMargin { get; set; }
public double OperatingMargin { get; set; }
public double NetIncome { get; set; }
public double Eps { get; set; }
public double TotalDebt { get; set; }
public double CashPosition { get; set; }
public int EmployeeCount { get; set; }
public List<string> KeyRisks { get; set; } = new();
public string Guidance { get; set; } = "";