How to correctly set Textract AnalyzeDocumentRequest syntax

[TheNopera]

I'm working on a demo project to extract information from medical exams using Amazon Textract.

Initially, I used the DetectDocumentText API and manually parsed the response to extract the values I needed. This approach worked in many cases, but it failed on others due to the complexity and layout variation of the documents.

Now, I'm trying to use Textract's Queries feature by passing a list of Query objects to AnalyzeDocument. However, I keep getting 500 Internal Server Error with the following error in Postman:

An error occurred: Request has invalid parameters
Here’s what I’ve already verified:

1. I checked the AWS SDK for .NET models to ensure I’m not missing any required parameters.
2. I confirmed that my list of Query objects is properly constructed and contains valid values.
3. I made sure the file I'm sending is a .png, and that it's below the size limits.
4. I confirmed that using a local file (byte array) instead of an S3 object is supported, according to this StackOverflow answer.

Here’s the relevant code from my service class:

using Amazon;
using Amazon.Runtime;
using Amazon.Textract;
using Amazon.Textract.Model;
using ExamReaderBackend.Models;

namespace ExamReaderBackend.Services;

public class TextractService
{
    private readonly AmazonTextractClient _textractClient;

    public TextractService()
    {
        var credentials = new BasicAWSCredentials("AWS_KEY", “AWS_SECRET_KEY”);

        _textractClient = new AmazonTextractClient(credentials, RegionEndpoint.USEast1);
    }

    public async Task<List<Block>> ExtractBlocksFromPdfAsync(IFormFile file)
    {
        // Read the file into a byte array first
        using var ms = new MemoryStream();
        await file.CopyToAsync(ms);
        var fileBytes = ms.ToArray();

        // Create a list of queries based on the exam targets
        var textractQueries = ExamTargets.List
        .Select(t => new Query
        {
            Text = t.Nome,
            Alias = t.Nome,
        }).ToList();
        Console.WriteLine($"Textract queries: {string.Join(", ", textractQueries.Select(q => q.Text))}”);

        // Create the AnalyzeDocumentRequest
        var request = new AnalyzeDocumentRequest
        {
            Document = new Document
            {
                Bytes = ms,
            },

            FeatureTypes = [FeatureType.QUERIES],
            QueriesConfig = new QueriesConfig
            {
                Queries = textractQueries
            },
        };

        Console.WriteLine("ENTERED TEXTRACT: “);

        var response = await _textractClient.AnalyzeDocumentAsync(request);

        Console.WriteLine("TEXTRACT RESPONSE RECEIVED”);

        if (response.HttpStatusCode != System.Net.HttpStatusCode.OK)
        {
            throw new Exception("Failed to analyze document.”);
        }
        return response.Blocks;
    }


}

[AlexDaines]

Hey there. Taking a look at your issue and I notice a potential problem. I think the issue may be with how you are handling the MemoryStream:

using var ms = new MemoryStream();
await file.CopyToAsync(ms);
var fileBytes = ms.ToArray();

// and then later on in your code

Document = new Document
{
    Bytes = ms, // This MemoryStream has already been read to the end
},

After calling ms.ToArray(), the MemoryStream's position is at the end of the stream. When you then use it for Document.Bytes, Textract sees an empty stream.

A couple solutions I see are:

// Reset the stream position after reading it
ms.Position = 0;

var request = new AnalyzeDocumentRequest
{
    Document = new Document
    {
        Bytes = ms,
    },
    // rest remains the same
};

Or, alternatively, you could just create a new MemoryStream from your byte array:

Document = new Document
{
    Bytes = new MemoryStream(fileBytes),
},

Could you give either of these a try and let us know if it resolves your issue?

Also! It's always reccomended to make use of AWS's credential providers rather than hard-coding your keys.

[TheNopera]

Oh I forgot to comment here how I fixed the issue, I did notice this mistake you pointed and fixed it afterwards. However the problem persisted. With the information I initially provided, no one would have been able to guess it (sorry about that!).

Root Cause

The issue was with the Text field in the Query objects I was sending to Textract:

Textract's Query.Text field does not accept special characters like accents (e.g., á, é, ã, etc.).
I was passing queries like "Exame de Próstata" or "Coração" which contain accented characters. Once I removed all accents from the Text values (e.g., changed "Próstata" to "Prostata"), everything worked fine. Thanks for the help

[github-actions[bot]]

Hello! Reopening this discussion to make it searchable.